Organizational	B-KEY
Self-Design	I-KEY
in	O
Semi-dynamic	O
Environments	O
ABSTRACT	O
In	O
this	O
paper	O
we	O
propose	O
a	O
run-time	O
approach	O
to	O
organization	B-KEY
that	O
is	O
contingent	O
on	O
the	O
task	O
structure	O
of	O
the	O
problem	O
being	O
solved	O
and	O
the	O
environmental	O
conditions	O
under	O
which	O
it	O
is	O
being	O
solved	O
.	O
We	O
use	O
T1EMS	O
as	O
the	O
underlying	O
representation	O
for	O
our	O
problems	O
and	O
describe	O
a	O
framework	O
that	O
uses	O
Organizational	B-KEY
Self-Design	I-KEY
-LRB-	O
OSD	O
-RRB-	O
to	O
allocate	O
tasks	O
and	O
resources	O
to	O
the	O
agents	O
and	O
coordinate	B-KEY
their	O
activities	O
.	O

The	O
Sequential	B-KEY
Auction	I-KEY
Problem	I-KEY
on	O
eBay	O
:	O
An	O
Empirical	B-KEY
Analysis	I-KEY
and	O
a	O
Solution	O
*	O
ABSTRACT	O
Bidders	O
on	O
eBay	O
have	O
no	O
dominant	O
bidding	B-KEY
strategy	I-KEY
when	O
faced	O
with	O
multiple	B-KEY
auctions	I-KEY
each	O
offering	O
an	O
item	O
of	O
interest	O
.	O
As	O
seen	O
through	O
an	O
analysis	O
of	O
1,956	O
auctions	O
on	O
eBay	O
for	O
a	O
Dell	O
E193FP	O
LCD	O
monitor	O
,	O
some	O
bidders	O
win	O
auctions	O
at	O
prices	O
higher	O
than	O
those	O
of	O
other	O
available	O
auctions	O
,	O
while	O
others	O
never	O
win	O
an	O
auction	O
despite	O
placing	O
bids	O
in	O
losing	O
efforts	O
that	O
are	O
greater	O
than	O
the	O
closing	O
prices	O
of	O
other	O
available	O
auctions	O
.	O
These	O
misqueues	O
in	O
strategic	B-KEY
behavior	I-KEY
hamper	O
the	O
efficiency	O
of	O
the	O
system	O
,	O
and	O
in	O
so	O
doing	O
limit	O
the	O
revenue	O
potential	O
for	O
sellers	O
.	O
This	O
paper	O
proposes	O
a	O
novel	O
options-based	O
extension	O
to	O
eBay	O
's	O
proxy-bidding	O
system	O
that	O
resolves	O
this	O
strategic	O
issue	O
for	O
buyers	O
in	O
commoditized	B-KEY
markets	I-KEY
.	O
An	O
empirical	B-KEY
analysis	I-KEY
of	O
eBay	O
provides	O
a	O
basis	O
for	O
computer	B-KEY
simulations	I-KEY
that	O
investigate	O
the	O
market	B-KEY
effects	I-KEY
of	O
the	O
options-based	O
scheme	O
,	O
and	O
demonstrates	O
that	O
the	O
options-based	O
scheme	O
provides	O
greater	O
efficiency	O
than	O
eBay	O
,	O
while	O
also	O
increasing	O
seller	O
revenue	O
.	O

Fairness	B-KEY
in	O
Dead-Reckoning	B-KEY
based	O
Distributed	O
Multi-Player	O
Games	O
ABSTRACT	O
In	O
a	O
distributed	O
multi-player	O
game	O
that	O
uses	O
dead-reckoning	B-KEY
vectors	O
to	O
exchange	O
movement	O
information	O
among	O
players	O
,	O
there	O
is	O
inaccuracy	O
in	O
rendering	O
the	O
objects	O
at	O
the	O
receiver	O
due	O
to	O
network	O
delay	O
between	O
the	O
sender	O
and	O
the	O
receiver	O
.	O
The	O
object	O
is	O
placed	O
at	O
the	O
receiver	O
at	O
the	O
position	O
indicated	O
by	O
the	O
dead-reckoning	B-KEY
vector	O
,	O
but	O
by	O
that	O
time	O
,	O
the	O
real	O
position	O
could	O
have	O
changed	O
considerably	O
at	O
the	O
sender	O
.	O
This	O
inaccuracy	O
would	O
be	O
tolerable	O
if	O
it	O
is	O
consistent	O
among	O
all	O
players	O
;	O
that	O
is	O
,	O
at	O
the	O
same	O
physical	O
time	O
,	O
all	O
players	O
see	O
inaccurate	O
-LRB-	O
with	O
respect	O
to	O
the	O
real	O
position	O
of	O
the	O
object	O
-RRB-	O
but	O
the	O
same	O
position	O
and	O
trajectory	O
for	O
an	O
object	O
.	O
But	O
due	O
to	O
varying	O
network	O
delays	O
between	O
the	O
sender	O
and	O
different	O
receivers	O
,	O
the	O
inaccuracy	O
is	O
different	O
at	O
different	O
players	O
as	O
well	O
.	O
This	O
leads	O
to	O
unfairness	O
in	O
game	O
playing	O
.	O
In	O
this	O
paper	O
,	O
we	O
first	O
introduce	O
an	O
``	O
error	O
''	O
measure	O
for	O
estimating	O
this	O
inaccuracy	O
.	O
Then	O
we	O
develop	O
an	O
algorithm	O
for	O
scheduling	O
the	O
sending	O
of	O
dead-reckoning	B-KEY
vectors	O
at	O
a	O
sender	O
that	O
strives	O
to	O
make	O
this	O
error	O
equal	O
at	O
different	O
receivers	O
over	O
time	O
.	O
This	O
algorithm	O
makes	O
the	O
game	O
very	O
fair	B-KEY
at	O
the	O
expense	O
of	O
increasing	O
the	O
overall	O
mean	B-KEY
error	I-KEY
of	O
all	O
players	O
.	O
To	O
mitigate	O
this	O
effect	O
,	O
we	O
propose	O
a	O
budget	B-KEY
based	I-KEY
algorithm	I-KEY
that	O
provides	O
improved	O
fairness	B-KEY
without	O
increasing	O
the	O
mean	B-KEY
error	I-KEY
thereby	O
maintaining	O
the	O
accuracy	B-KEY
of	O
game	O
playing	O
.	O
We	O
have	O
implemented	O
both	O
the	O
scheduling	B-KEY
algorithm	I-KEY
and	O
the	O
budget	B-KEY
based	I-KEY
algorithm	I-KEY
as	O
part	O
of	O
BZFlag	O
,	O
a	O
popular	O
distributed	O
multi-player	O
game	O
.	O
We	O
show	O
through	O
experiments	O
that	O
these	O
algorithms	O
provide	O
fairness	B-KEY
among	O
players	O
in	O
spite	O
of	O
widely	O
varying	O
network	O
delays	O
.	O
An	O
additional	O
property	O
of	O
the	O
proposed	O
algorithms	O
is	O
that	O
they	O
require	O
less	O
number	O
of	O
DRs	O
to	O
be	O
exchanged	O
-LRB-	O
compared	O
to	O
the	O
current	O
implementation	O
of	O
BZflag	O
-RRB-	O
to	O
achieve	O
the	O
same	O
level	O
of	O
accuracy	B-KEY
in	O
game	O
playing	O
.	O

Negotiation-Range	O
Mechanisms	O
:	O
Exploring	O
the	O
Limits	O
of	O
Truthful	O
Efficient	B-KEY
Markets	I-KEY
ABSTRACT	O
This	O
paper	O
introduces	O
a	O
new	O
class	O
of	O
mechanisms	O
based	O
on	O
negotiation	O
between	O
market	O
participants	O
.	O
This	O
model	O
allows	O
us	O
to	O
circumvent	O
Myerson	O
and	O
Satterthwaite	O
's	O
impossibility	B-KEY
result	I-KEY
and	O
present	O
a	O
bilateral	O
market	O
mechanism	O
that	O
is	O
efficient	O
,	O
individually	B-KEY
rational	I-KEY
,	O
incentive	B-KEY
compatible	I-KEY
and	O
budget	O
balanced	O
in	O
the	O
single-unit	O
heterogeneous	O
setting	O
.	O
The	O
underlying	O
scheme	O
makes	O
this	O
combination	O
of	O
desirable	O
qualities	O
possible	O
by	O
reporting	O
a	O
price	O
range	O
for	O
each	O
buyer-seller	O
pair	O
that	O
defines	O
a	O
zone	B-KEY
of	I-KEY
possible	I-KEY
agreements	I-KEY
,	O
while	O
the	O
final	O
price	O
is	O
left	O
open	O
for	O
negotiation	O
.	O

Cost	B-KEY
Sharing	I-KEY
in	O
a	O
Job	B-KEY
Scheduling	I-KEY
Problem	O
Using	O
the	O
Shapley	O
Value	O
ABSTRACT	O
A	O
set	O
of	O
jobs	O
need	O
to	O
be	O
served	O
by	O
a	O
single	O
server	O
which	O
can	O
serve	O
only	O
one	O
job	O
at	O
a	O
time	O
.	O
Jobs	O
have	O
processing	B-KEY
times	I-KEY
and	O
incur	O
waiting	O
costs	O
-LRB-	O
linear	O
in	O
their	O
waiting	O
time	O
-RRB-	O
.	O
The	O
jobs	O
share	O
their	O
costs	O
through	O
compensation	O
using	O
monetary	B-KEY
transfers	I-KEY
.	O
We	O
characterize	O
the	O
Shapley	O
value	O
rule	O
for	O
this	O
model	O
using	O
fairness	B-KEY
axioms	I-KEY
.	O
Our	O
axioms	O
include	O
a	O
bound	O
on	O
the	O
cost	B-KEY
share	I-KEY
of	O
jobs	O
in	O
a	O
group	O
,	O
efficiency	O
,	O
and	O
some	O
independence	O
properties	O
on	O
the	O
the	O
cost	B-KEY
share	I-KEY
of	O
a	O
job	O
.	O

Combinatorial	O
Resource	B-KEY
Scheduling	O
for	O
Multiagent	O
MDPs	O
ABSTRACT	O
Optimal	O
resource	B-KEY
scheduling	O
in	O
multiagent	O
systems	O
is	O
a	O
computationally	O
challenging	O
task	O
,	O
particularly	O
when	O
the	O
values	O
of	O
resources	O
are	O
not	O
additive	O
.	O
We	O
consider	O
the	O
combinatorial	O
problem	O
of	O
scheduling	B-KEY
the	O
usage	O
of	O
multiple	O
resources	B-KEY
among	O
agents	O
that	O
operate	O
in	O
stochastic	O
environments	O
,	O
modeled	O
as	O
Markov	B-KEY
decision	I-KEY
processes	I-KEY
-LRB-	O
MDPs	O
-RRB-	O
.	O
In	O
recent	O
years	O
,	O
efficient	O
resource-allocation	O
algorithms	O
have	O
been	O
developed	O
for	O
agents	O
with	O
resource	B-KEY
values	O
induced	O
by	O
MDPs	O
.	O
However	O
,	O
this	O
prior	O
work	O
has	O
focused	O
on	O
static	O
resource-allocation	O
problems	O
where	O
resources	B-KEY
are	O
distributed	O
once	O
and	O
then	O
utilized	O
in	O
infinite-horizon	O
MDPs	O
.	O
We	O
extend	O
those	O
existing	O
models	O
to	O
the	O
problem	O
of	O
combinatorial	O
resource	B-KEY
scheduling	O
,	O
where	O
agents	O
persist	O
only	O
for	O
finite	O
periods	O
between	O
their	O
-LRB-	O
predefined	O
-RRB-	O
arrival	O
and	O
departure	O
times	O
,	O
requiring	O
resources	O
only	O
for	O
those	O
time	O
periods	O
.	O
We	O
provide	O
a	O
computationally	O
efficient	O
procedure	O
for	O
computing	O
globally	O
optimal	O
resource	B-KEY
assignments	O
to	O
agents	O
over	O
time	O
.	O
We	O
illustrate	O
and	O
empirically	O
analyze	O
the	O
method	O
in	O
the	O
context	O
of	O
a	O
stochastic	O
jobscheduling	O
domain	O
.	O

Estimating	O
the	O
Global	B-KEY
PageRank	I-KEY
of	O
Web	B-KEY
Communities	I-KEY
ABSTRACT	O
Localized	B-KEY
search	I-KEY
engines	I-KEY
are	O
small-scale	O
systems	O
that	O
index	O
a	O
particular	O
community	O
on	O
the	O
web	O
.	O
They	O
offer	O
several	O
benefits	O
over	O
their	O
large-scale	O
counterparts	O
in	O
that	O
they	O
are	O
relatively	O
inexpensive	O
to	O
build	O
,	O
and	O
can	O
provide	O
more	O
precise	O
and	O
complete	O
search	O
capability	O
over	O
their	O
relevant	O
domains	O
.	O
One	O
disadvantage	O
such	O
systems	O
have	O
over	O
large-scale	O
search	O
engines	O
is	O
the	O
lack	O
of	O
global	B-KEY
PageRank	I-KEY
values	O
.	O
Such	O
information	O
is	O
needed	O
to	O
assess	O
the	O
value	O
of	O
pages	O
in	O
the	O
localized	O
search	O
domain	O
within	O
the	O
context	O
of	O
the	O
web	O
as	O
a	O
whole	O
.	O
In	O
this	O
paper	O
,	O
we	O
present	O
well-motivated	O
algorithms	B-KEY
to	O
estimate	O
the	O
global	B-KEY
PageRank	I-KEY
values	O
of	O
a	O
local	B-KEY
domain	I-KEY
.	O
The	O
algorithms	B-KEY
are	O
all	O
highly	O
scalable	O
in	O
that	O
,	O
given	O
a	O
local	B-KEY
domain	I-KEY
of	O
size	O
n	O
,	O
they	O
use	O
O	O
-LRB-	O
n	O
-RRB-	O
resources	O
that	O
include	O
computation	O
time	O
,	O
bandwidth	O
,	O
and	O
storage	O
.	O
We	O
test	O
our	O
methods	O
across	O
a	O
variety	O
of	O
localized	B-KEY
domains	I-KEY
,	O
including	O
site-specific	O
domains	O
and	O
topic-specific	B-KEY
domains	I-KEY
.	O
We	O
demonstrate	O
that	O
by	O
crawling	O
as	O
few	O
as	O
n	O
or	O
2n	O
additional	O
pages	O
,	O
our	O
methods	O
can	O
give	O
excellent	O
global	B-KEY
PageRank	I-KEY
estimates	O
.	O

On	O
the	O
Benefits	O
of	O
Cheating	O
by	O
Self-Interested	B-KEY
Agents	I-KEY
in	O
Vehicular	B-KEY
Networks	I-KEY
∗	O
ABSTRACT	O
As	O
more	O
and	O
more	O
cars	O
are	O
equipped	O
with	O
GPS	O
and	O
Wi-Fi	O
transmitters	O
,	O
it	O
becomes	O
easier	O
to	O
design	O
systems	O
that	O
will	O
allow	O
cars	O
to	O
interact	O
autonomously	O
with	O
each	O
other	O
,	O
e.g.	O
,	O
regarding	O
traffic	O
on	O
the	O
roads	O
.	O
Indeed	O
,	O
car	O
manufacturers	O
are	O
already	O
equipping	O
their	O
cars	O
with	O
such	O
devices	O
.	O
Though	O
,	O
currently	O
these	O
systems	O
are	O
a	O
proprietary	O
,	O
we	O
envision	O
a	O
natural	O
evolution	O
where	O
agent	O
applications	O
will	O
be	O
developed	O
for	O
vehicular	O
systems	O
,	O
e.g.	O
,	O
to	O
improve	O
car	O
routing	O
in	O
dense	O
urban	O
areas	O
.	O
Nonetheless	O
,	O
this	O
new	O
technology	O
and	O
agent	O
applications	O
may	O
lead	O
to	O
the	O
emergence	O
of	O
self-interested	O
car	O
owners	O
,	O
who	O
will	O
care	O
more	O
about	O
their	O
own	O
welfare	O
than	O
the	O
social	O
welfare	O
of	O
their	O
peers	O
.	O
These	O
car	O
owners	O
will	O
try	O
to	O
manipulate	O
their	O
agents	O
such	O
that	O
they	O
transmit	O
false	O
data	O
to	O
their	O
peers	O
.	O
Using	O
a	O
simulation	O
environment	O
,	O
which	O
models	O
a	O
real	O
transportation	O
network	O
in	O
a	O
large	O
city	O
,	O
we	O
demonstrate	O
the	O
benefits	O
achieved	O
by	O
self-interested	B-KEY
agents	I-KEY
if	O
no	O
counter-measures	O
are	O
implemented	O
.	O

MSP	O
:	O
Multi-Sequence	O
Positioning	O
of	O
Wireless	O
Sensor	O
Nodes	O
*	O
Abstract	O
Wireless	B-KEY
Sensor	I-KEY
Networks	I-KEY
have	O
been	O
proposed	O
for	O
use	O
in	O
many	O
location-dependent	O
applications	O
.	O
Most	O
of	O
these	O
need	O
to	O
identify	O
the	O
locations	O
of	O
wireless	O
sensor	O
nodes	O
,	O
a	O
challenging	O
task	O
because	O
of	O
the	O
severe	O
constraints	O
on	O
cost	O
,	O
energy	O
and	O
effective	O
range	O
of	O
sensor	O
devices	O
.	O
To	O
overcome	O
limitations	O
in	O
existing	O
solutions	O
,	O
we	O
present	O
a	O
Multi-Sequence	O
Positioning	O
-LRB-	O
MSP	O
-RRB-	O
method	O
for	O
large-scale	O
stationary	O
sensor	O
node	B-KEY
localization	I-KEY
in	O
outdoor	O
environments	O
.	O
The	O
novel	O
idea	O
behind	O
MSP	O
is	O
to	O
reconstruct	O
and	O
estimate	O
two-dimensional	O
location	O
information	O
for	O
each	O
sensor	O
node	O
by	O
processing	O
multiple	O
one-dimensional	O
node	O
sequences	O
,	O
easily	O
obtained	O
through	O
loosely	O
guided	O
event	B-KEY
distribution	I-KEY
.	O
Starting	O
from	O
a	O
basic	O
MSP	O
design	O
,	O
we	O
propose	O
four	O
optimizations	O
,	O
which	O
work	O
together	O
to	O
increase	O
the	O
localization	B-KEY
accuracy	O
.	O
We	O
address	O
several	O
interesting	O
issues	O
,	O
such	O
as	O
incomplete	O
-LRB-	O
partial	O
-RRB-	O
node	O
sequences	O
and	O
sequence	O
flip	O
,	O
found	O
in	O
the	O
Mirage	O
test-bed	O
we	O
built	O
.	O
We	O
have	O
evaluated	O
the	O
MSP	O
system	O
through	O
theoretical	O
analysis	O
,	O
extensive	O
simulation	O
as	O
well	O
as	O
two	O
physical	O
systems	O
-LRB-	O
an	O
indoor	O
version	O
with	O
46	O
MICAz	O
motes	O
and	O
an	O
outdoor	O
version	O
with	O
20	O
MICAz	O
motes	O
-RRB-	O
.	O
This	O
evaluation	O
demonstrates	O
that	O
MSP	O
can	O
achieve	O
an	O
accuracy	O
within	O
one	O
foot	O
,	O
requiring	O
neither	O
additional	O
costly	O
hardware	O
on	O
sensor	O
nodes	O
nor	O
precise	O
event	B-KEY
distribution	I-KEY
.	O
It	O
also	O
provides	O
a	O
nice	O
tradeoff	O
between	O
physical	O
cost	O
-LRB-	O
anchors	O
-RRB-	O
and	O
soft	O
cost	O
-LRB-	O
events	O
-RRB-	O
,	O
while	O
maintaining	O
localization	B-KEY
accuracy	O
.	O

Letting	O
loose	O
a	O
SPIDER	O
on	O
a	O
network	B-KEY
of	O
POMDPs	B-KEY
:	O
Generating	O
quality	B-KEY
guaranteed	I-KEY
policies	I-KEY
ABSTRACT	O
Distributed	B-KEY
Partially	I-KEY
Observable	I-KEY
Markov	I-KEY
Decision	I-KEY
Problems	I-KEY
-LRB-	O
Distributed	B-KEY
POMDPs	I-KEY
-RRB-	O
are	O
a	O
popular	O
approach	O
for	O
modeling	O
multi-agent	O
systems	O
acting	O
in	O
uncertain	O
domains	O
.	O
Given	O
the	O
significant	O
complexity	O
of	O
solving	O
distributed	B-KEY
POMDPs	I-KEY
,	O
particularly	O
as	O
we	O
scale	O
up	O
the	O
numbers	O
of	O
agents	O
,	O
one	O
popular	O
approach	O
has	O
focused	O
on	O
approximate	O
solutions	O
.	O
Though	O
this	O
approach	O
is	O
efficient	O
,	O
the	O
algorithms	O
within	O
this	O
approach	O
do	O
not	O
provide	O
any	O
guarantees	O
on	O
solution	O
quality	O
.	O
A	O
second	O
less	O
popular	O
approach	O
focuses	O
on	O
global	B-KEY
optimality	I-KEY
,	O
but	O
typical	O
results	O
are	O
available	O
only	O
for	O
two	O
agents	O
,	O
and	O
also	O
at	O
considerable	O
computational	O
cost	O
.	O
This	O
paper	O
overcomes	O
the	O
limitations	O
of	O
both	O
these	O
approaches	O
by	O
providing	O
SPIDER	O
,	O
a	O
novel	O
combination	O
of	O
three	O
key	O
features	O
for	O
policy	O
generation	O
in	O
distributed	B-KEY
POMDPs	I-KEY
:	O
-LRB-	O
i	O
-RRB-	O
it	O
exploits	O
agent	O
interaction	O
structure	O
given	O
a	O
network	O
of	O
agents	O
-LRB-	O
i.e.	O
allowing	O
easier	O
scale-up	O
to	O
larger	O
number	O
of	O
agents	O
-RRB-	O
;	O
-LRB-	O
ii	O
-RRB-	O
it	O
uses	O
a	O
combination	O
of	O
heuristics	O
to	O
speedup	O
policy	O
search	O
;	O
and	O
-LRB-	O
iii	O
-RRB-	O
it	O
allows	O
quality	O
guaranteed	O
approximations	O
,	O
allowing	O
a	O
systematic	O
tradeoff	O
of	O
solution	O
quality	O
for	O
time	O
.	O
Experimental	O
results	O
show	O
orders	O
of	O
magnitude	O
improvement	O
in	O
performance	O
when	O
compared	O
with	O
previous	O
global	B-KEY
optimal	I-KEY
algorithms	O
.	O

A	O
Point-Distribution	B-KEY
Index	I-KEY
and	O
Its	O
Application	O
to	O
Sensor-Grouping	B-KEY
in	O
Wireless	B-KEY
Sensor	I-KEY
Networks	I-KEY
ABSTRACT	O
We	O
propose	O
t	O
,	O
a	O
novel	O
index	O
for	O
evaluation	O
of	O
point-distribution	O
.	O
t	O
is	O
the	O
minimum	O
distance	O
between	O
each	O
pair	O
of	O
points	O
normalized	O
by	O
the	O
average	O
distance	O
between	O
each	O
pair	O
of	O
points	O
.	O
We	O
find	O
that	O
a	O
set	O
of	O
points	O
that	O
achieve	O
a	O
maximum	O
value	O
of	O
t	O
result	O
in	O
a	O
honeycomb	B-KEY
structure	I-KEY
.	O
We	O
propose	O
that	O
t	O
can	O
serve	O
as	O
a	O
good	O
index	O
to	O
evaluate	O
the	O
distribution	O
of	O
the	O
points	O
,	O
which	O
can	O
be	O
employed	O
in	O
coverage-related	O
problems	O
in	O
wireless	B-KEY
sensor	I-KEY
networks	I-KEY
-LRB-	O
WSNs	O
-RRB-	O
.	O
To	O
validate	O
this	O
idea	O
,	O
we	O
formulate	O
a	O
general	O
sensorgrouping	O
problem	O
for	O
WSNs	O
and	O
provide	O
a	O
general	O
sensing	O
model	O
.	O
We	O
show	O
that	O
locally	O
maximizing	O
t	O
at	O
sensor	O
nodes	O
is	O
a	O
good	O
approach	O
to	O
solve	O
this	O
problem	O
with	O
an	O
algorithm	O
called	O
Maximizingt	O
Node-Deduction	O
-LRB-	O
MIND	O
-RRB-	O
.	O
Simulation	O
results	O
verify	O
that	O
MIND	O
outperforms	O
a	O
greedy	O
algorithm	O
that	O
exploits	O
sensor-redundancy	O
we	O
design	O
.	O
This	O
demonstrates	O
a	O
good	O
application	O
of	O
employing	O
t	O
in	O
coverage-related	O
problems	O
for	O
WSNs	O
.	O

Robust	O
Incentive	B-KEY
Techniques	O
for	O
Peer-to-Peer	O
Networks	O
Lack	O
of	O
cooperation	O
-LRB-	O
free	O
riding	O
-RRB-	O
is	O
one	O
of	O
the	O
key	O
problems	O
that	O
confronts	O
today	O
's	O
P2P	B-KEY
systems	I-KEY
.	O
What	O
makes	O
this	O
problem	O
particularly	O
difficult	O
is	O
the	O
unique	O
set	O
of	O
challenges	O
that	O
P2P	B-KEY
systems	I-KEY
pose	O
:	O
large	O
populations	O
,	O
high	O
turnover	O
,	O
asymmetry	O
of	O
interest	O
,	O
collusion	B-KEY
,	O
zero-cost	O
identities	O
,	O
and	O
traitors	O
.	O
To	O
tackle	O
these	O
challenges	O
we	O
model	O
the	O
P2P	B-KEY
system	I-KEY
using	O
the	O
Generalized	O
Prisoner	O
's	O
Dilemma	O
-LRB-	O
GPD	O
-RRB-	O
,	O
and	O
propose	O
the	O
Reciprocative	B-KEY
decision	I-KEY
function	I-KEY
as	O
the	O
basis	O
of	O
a	O
family	O
of	O
incentives	B-KEY
techniques	O
.	O
These	O
techniques	O
are	O
fully	O
distributed	O
and	O
include	O
:	O
discriminating	O
server	O
selection	O
,	O
maxflowbased	O
subjective	O
reputation	B-KEY
,	O
and	O
adaptive	B-KEY
stranger	I-KEY
policies	I-KEY
.	O
Through	O
simulation	O
,	O
we	O
show	O
that	O
these	O
techniques	O
can	O
drive	O
a	O
system	O
of	O
strategic	O
users	O
to	O
nearly	O
optimal	O
levels	O
of	O
cooperation	O
.	O

A	O
Time	B-KEY
Machine	I-KEY
for	O
Text	B-KEY
Search	I-KEY
ABSTRACT	O
Text	B-KEY
search	I-KEY
over	O
temporally	O
versioned	B-KEY
document	I-KEY
collections	I-KEY
such	O
as	O
web	B-KEY
archives	I-KEY
has	O
received	O
little	O
attention	O
as	O
a	O
research	O
problem	O
.	O
As	O
a	O
consequence	O
,	O
there	O
is	O
no	O
scalable	O
and	O
principled	O
solution	O
to	O
search	O
such	O
a	O
collection	O
as	O
of	O
a	O
specified	O
time	O
t	O
.	O
In	O
this	O
work	O
,	O
we	O
address	O
this	O
shortcoming	O
and	O
propose	O
an	O
efficient	O
solution	O
for	O
time-travel	O
text	B-KEY
search	I-KEY
by	O
extending	O
the	O
inverted	O
file	O
index	O
to	O
make	O
it	O
ready	O
for	O
temporal	O
search	O
.	O
We	O
introduce	O
approximate	B-KEY
temporal	I-KEY
coalescing	I-KEY
as	O
a	O
tunable	O
method	O
to	O
reduce	O
the	O
index	O
size	O
without	O
significantly	O
affecting	O
the	O
quality	O
of	O
results	O
.	O
In	O
order	O
to	O
further	O
improve	O
the	O
performance	O
of	O
time-travel	O
queries	O
,	O
we	O
introduce	O
two	O
principled	O
techniques	O
to	O
trade	O
off	O
index	O
size	O
for	O
its	O
performance	O
.	O
These	O
techniques	O
can	O
be	O
formulated	O
as	O
optimization	O
problems	O
that	O
can	O
be	O
solved	O
to	O
near-optimality	O
.	O
Finally	O
,	O
our	O
approach	O
is	O
evaluated	O
in	O
a	O
comprehensive	O
series	O
of	O
experiments	O
on	O
two	O
large-scale	O
real-world	O
datasets	O
.	O
Results	O
unequivocally	O
show	O
that	O
our	O
methods	O
make	O
it	O
possible	O
to	O
build	O
an	O
efficient	O
``	O
time	B-KEY
machine	I-KEY
''	O
scalable	O
to	O
large	O
versioned	O
text	O
collections	O
.	O

pTHINC	B-KEY
:	O
A	O
Thin-Client	B-KEY
Architecture	O
for	O
Mobile	B-KEY
Wireless	O
Web	O
ABSTRACT	O
Although	O
web	B-KEY
applications	I-KEY
are	O
gaining	O
popularity	O
on	O
mobile	B-KEY
wireless	O
PDAs	O
,	O
web	O
browsers	O
on	O
these	O
systems	O
can	O
be	O
quite	O
slow	O
and	O
often	O
lack	O
adequate	O
functionality	O
to	O
access	O
many	O
web	O
sites	O
.	O
We	O
have	O
developed	O
pTHINC	B-KEY
,	O
a	O
PDA	B-KEY
thinclient	I-KEY
solution	I-KEY
that	O
leverages	O
more	O
powerful	O
servers	O
to	O
run	O
full-function	O
web	B-KEY
browsers	I-KEY
and	O
other	O
application	O
logic	O
,	O
then	O
sends	O
simple	O
screen	O
updates	O
to	O
the	O
PDA	O
for	O
display	O
.	O
pTHINC	B-KEY
uses	O
server-side	O
screen	O
scaling	O
to	O
provide	O
high-fidelity	O
display	O
and	O
seamless	B-KEY
mobility	I-KEY
across	O
a	O
broad	O
range	O
of	O
different	O
clients	O
and	O
screen	O
sizes	O
,	O
including	O
both	O
portrait	O
and	O
landscape	O
viewing	O
modes	O
.	O
pTHINC	B-KEY
also	O
leverages	O
existing	O
PDA	O
control	O
buttons	O
to	O
improve	O
system	B-KEY
usability	I-KEY
and	O
maximize	O
available	O
screen	B-KEY
resolution	I-KEY
for	O
application	O
display	O
.	O
We	O
have	O
implemented	O
pTHINC	B-KEY
on	O
Windows	O
Mobile	B-KEY
and	O
evaluated	O
its	O
performance	O
on	O
mobile	B-KEY
wireless	O
devices	O
.	O
Our	O
results	O
compared	O
to	O
local	B-KEY
PDA	I-KEY
web	I-KEY
browsers	I-KEY
and	O
other	O
thin-client	O
approaches	O
demonstrate	O
that	O
pTHINC	O
provides	O
superior	O
web	O
browsing	O
performance	O
and	O
is	O
the	O
only	O
PDA	O
thin	O
client	O
that	O
effectively	O
supports	O
crucial	O
browser	O
helper	O
applications	O
such	O
as	O
video	O
playback	O
.	O
Categories	O
and	O
Subject	O
Descriptors	O
:	O
C.	O
2.4	O
ComputerCommunication-Networks	O
:	O
Distributed	O
Systems	O
--	O
client	O
/	O
server	O

Graphical	B-KEY
Models	I-KEY
for	O
Online	O
Solutions	O
to	O
Interactive	O
POMDPs	O
ABSTRACT	O
We	O
develop	O
a	O
new	O
graphical	O
representation	O
for	O
interactive	B-KEY
partially	I-KEY
observable	I-KEY
Markov	I-KEY
decision	I-KEY
processes	I-KEY
-LRB-	O
I-POMDPs	O
-RRB-	O
that	O
is	O
significantly	O
more	O
transparent	O
and	O
semantically	O
clear	O
than	O
the	O
previous	O
representation	O
.	O
These	O
graphical	B-KEY
models	I-KEY
called	O
interactive	B-KEY
dynamic	I-KEY
influence	I-KEY
diagrams	I-KEY
-LRB-	O
I-DIDs	O
-RRB-	O
seek	O
to	O
explicitly	O
model	O
the	O
structure	O
that	O
is	O
often	O
present	O
in	O
real-world	O
problems	O
by	O
decomposing	O
the	O
situation	O
into	O
chance	O
and	O
decision	O
variables	O
,	O
and	O
the	O
dependencies	O
between	O
the	O
variables	O
.	O
I-DIDs	O
generalize	O
DIDs	O
,	O
which	O
may	O
be	O
viewed	O
as	O
graphical	O
representations	O
of	O
POMDPs	O
,	O
to	O
multiagent	O
settings	O
in	O
the	O
same	O
way	O
that	O
I-POMDPs	O
generalize	O
POMDPs	O
.	O
I-DIDs	O
may	O
be	O
used	O
to	O
compute	O
the	O
policy	O
of	O
an	O
agent	B-KEY
online	I-KEY
as	O
the	O
agent	O
acts	O
and	O
observes	O
in	O
a	O
setting	O
that	O
is	O
populated	O
by	O
other	O
interacting	O
agents	O
.	O
Using	O
several	O
examples	O
,	O
we	O
show	O
how	O
I-DIDs	O
may	O
be	O
applied	O
and	O
demonstrate	O
their	O
usefulness	O
.	O

An	O
Outranking	B-KEY
Approach	I-KEY
for	O
Rank	B-KEY
Aggregation	I-KEY
in	O
Information	B-KEY
Retrieval	I-KEY
ABSTRACT	O
Research	O
in	O
Information	B-KEY
Retrieval	I-KEY
usually	O
shows	O
performance	O
improvement	O
when	O
many	O
sources	O
of	O
evidence	O
are	O
combined	O
to	O
produce	O
a	O
ranking	O
of	O
documents	O
-LRB-	O
e.g.	O
,	O
texts	O
,	O
pictures	O
,	O
sounds	O
,	O
etc.	O
-RRB-	O
.	O
In	O
this	O
paper	O
,	O
we	O
focus	O
on	O
the	O
rank	B-KEY
aggregation	I-KEY
problem	O
,	O
also	O
called	O
data	B-KEY
fusion	I-KEY
problem	O
,	O
where	O
rankings	O
of	O
documents	O
,	O
searched	O
into	O
the	O
same	O
collection	O
and	O
provided	O
by	O
multiple	O
methods	O
,	O
are	O
combined	O
in	O
order	O
to	O
produce	O
a	O
new	O
ranking	O
.	O
In	O
this	O
context	O
,	O
we	O
propose	O
a	O
rank	B-KEY
aggregation	I-KEY
method	O
within	O
a	O
multiple	O
criteria	O
framework	O
using	O
aggregation	O
mechanisms	O
based	O
on	O
decision	B-KEY
rules	I-KEY
identifying	O
positive	O
and	O
negative	O
reasons	O
for	O
judging	O
whether	O
a	O
document	O
should	O
get	O
a	O
better	O
rank	O
than	O
another	O
.	O
We	O
show	O
that	O
the	O
proposed	O
method	O
deals	O
well	O
with	O
the	O
Information	B-KEY
Retrieval	I-KEY
distinctive	O
features	O
.	O
Experimental	O
results	O
are	O
reported	O
showing	O
that	O
the	O
suggested	O
method	O
performs	O
better	O
than	O
the	O
well-known	O
CombSUM	O
and	O
CombMNZ	O
operators	O
.	O

Multi-dimensional	B-KEY
Range	I-KEY
Queries	I-KEY
in	O
Sensor	B-KEY
Networks	I-KEY
*	O
ABSTRACT	O
In	O
many	O
sensor	B-KEY
networks	I-KEY
,	O
data	O
or	O
events	O
are	O
named	O
by	O
attributes	O
.	O
Many	O
of	O
these	O
attributes	O
have	O
scalar	O
values	O
,	O
so	O
one	O
natural	O
way	O
to	O
query	O
events	O
of	O
interest	O
is	O
to	O
use	O
a	O
multidimensional	B-KEY
range	I-KEY
query	I-KEY
.	O
An	O
example	O
is	O
:	O
``	O
List	O
all	O
events	O
whose	O
temperature	O
lies	O
between	O
50	O
◦	O
and	O
60	O
◦	O
,	O
and	O
whose	O
light	O
levels	O
lie	O
between	O
10	O
and	O
15	O
.	O
''	O
Such	O
queries	O
are	O
useful	O
for	O
correlating	O
events	O
occurring	O
within	O
the	O
network	O
.	O
In	O
this	O
paper	O
,	O
we	O
describe	O
the	O
design	O
of	O
a	O
distributed	B-KEY
index	I-KEY
that	O
scalably	O
supports	O
multi-dimensional	B-KEY
range	I-KEY
queries	I-KEY
.	O
Our	O
distributed	B-KEY
index	I-KEY
for	O
multi-dimensional	O
data	O
-LRB-	O
or	O
DIM	B-KEY
-RRB-	O
uses	O
a	O
novel	O
geographic	O
embedding	O
of	O
a	O
classical	O
index	O
data	O
structure	O
,	O
and	O
is	O
built	O
upon	O
the	O
GPSR	O
geographic	B-KEY
routing	I-KEY
algorithm	O
.	O
Our	O
analysis	O
reveals	O
that	O
,	O
under	O
reasonable	O
assumptions	O
about	O
query	O
distributions	O
,	O
DIMs	B-KEY
scale	O
quite	O
well	O
with	O
network	O
size	O
-LRB-	O
both	O
insertion	O
and	O
query	B-KEY
costs	I-KEY
scale	O
as	O
O	O
-LRB-	O
√	O
N	O
-RRB-	O
-RRB-	O
.	O
In	O
detailed	O
simulations	O
,	O
we	O
show	O
that	O
in	O
practice	O
,	O
the	O
insertion	O
and	O
query	B-KEY
costs	I-KEY
of	O
other	O
alternatives	O
are	O
sometimes	O
an	O
order	O
of	O
magnitude	O
more	O
than	O
the	O
costs	O
of	O
DIMs	B-KEY
,	O
even	O
for	O
moderately	O
sized	O
network	O
.	O
Finally	O
,	O
experiments	O
on	O
a	O
small	O
scale	O
testbed	O
validate	O
the	O
feasibility	O
of	O
DIMs	B-KEY
.	O

Tracking	O
Immediate	B-KEY
Predecessors	I-KEY
in	O
Distributed	B-KEY
Computations	I-KEY
ABSTRACT	O
A	O
distributed	B-KEY
computation	I-KEY
is	O
usually	O
modeled	O
as	O
a	O
partially	O
ordered	O
set	O
of	O
relevant	B-KEY
events	I-KEY
-LRB-	O
the	O
relevant	B-KEY
events	I-KEY
are	O
a	O
subset	O
of	O
the	O
primitive	O
events	O
produced	O
by	O
the	O
computation	O
-RRB-	O
.	O
An	O
important	O
causality-related	O
distributed	B-KEY
computing	I-KEY
problem	O
,	O
that	O
we	O
call	O
the	O
Immediate	B-KEY
Predecessors	I-KEY
Tracking	O
-LRB-	O
IPT	O
-RRB-	O
problem	O
,	O
consists	O
in	O
associating	O
with	O
each	O
relevant	O
event	O
,	O
on	O
the	O
fly	O
and	O
without	O
using	O
additional	O
control	O
messages	O
,	O
the	O
set	O
of	O
relevant	O
events	O
that	O
are	O
its	O
immediate	O
predecessors	O
in	O
the	O
partial	O
order	O
.	O
So	O
,	O
IPT	O
is	O
the	O
on-the-fly	O
computation	O
of	O
the	O
transitive	B-KEY
reduction	I-KEY
-LRB-	O
i.e.	O
,	O
Hasse	B-KEY
diagram	I-KEY
-RRB-	O
of	O
the	O
causality	O
relation	O
defined	O
by	O
a	O
distributed	B-KEY
computation	I-KEY
.	O
This	O
paper	O
addresses	O
the	O
IPT	O
problem	O
:	O
it	O
presents	O
a	O
family	O
of	O
protocols	O
that	O
provides	O
each	O
relevant	B-KEY
event	I-KEY
with	O
a	O
timestamp	B-KEY
that	O
exactly	O
identifies	O
its	O
immediate	B-KEY
predecessors	I-KEY
.	O
The	O
family	O
is	O
defined	O
by	O
a	O
general	O
condition	O
that	O
allows	O
application	O
messages	O
to	O
piggyback	B-KEY
control	B-KEY
information	I-KEY
whose	O
size	O
can	O
be	O
smaller	O
than	O
n	O
-LRB-	O
the	O
number	O
of	O
processes	O
-RRB-	O
.	O
In	O
that	O
sense	O
,	O
this	O
family	O
defines	O
message	O
size-efficient	O
IPT	B-KEY
protocols	I-KEY
.	O
According	O
to	O
the	O
way	O
the	O
general	O
condition	O
is	O
implemented	O
,	O
different	O
IPT	B-KEY
protocols	I-KEY
can	O
be	O
obtained	O
.	O
Two	O
of	O
them	O
are	O
exhibited	O
.	O

Ranking	B-KEY
Web	B-KEY
Objects	I-KEY
from	O
Multiple	O
Communities	O
ABSTRACT	O
Vertical	B-KEY
search	I-KEY
is	O
a	O
promising	O
direction	O
as	O
it	O
leverages	O
domainspecific	O
knowledge	O
and	O
can	O
provide	O
more	O
precise	O
information	O
for	O
users	O
.	O
In	O
this	O
paper	O
,	O
we	O
study	O
the	O
Web	O
object-ranking	O
problem	O
,	O
one	O
of	O
the	O
key	O
issues	O
in	O
building	O
a	O
vertical	O
search	O
engine	O
.	O
More	O
specifically	O
,	O
we	O
focus	O
on	O
this	O
problem	O
in	O
cases	O
when	O
objects	O
lack	O
relationships	O
between	O
different	O
Web	O
communities	O
,	O
and	O
take	O
high-quality	O
photo	O
search	O
as	O
the	O
test	O
bed	O
for	O
this	O
investigation	O
.	O
We	O
proposed	O
two	O
score	B-KEY
fusion	I-KEY
methods	I-KEY
that	O
can	O
automatically	O
integrate	O
as	O
many	O
Web	O
communities	O
-LRB-	O
Web	O
forums	O
-RRB-	O
with	O
rating	O
information	O
as	O
possible	O
.	O
The	O
proposed	O
fusion	O
methods	O
leverage	O
the	O
hidden	O
links	O
discovered	O
by	O
a	O
duplicate	B-KEY
photo	I-KEY
detection	I-KEY
algorithm	I-KEY
,	O
and	O
aims	O
at	O
minimizing	O
score	O
differences	O
of	O
duplicate	O
photos	O
in	O
different	O
forums	O
.	O
Both	O
intermediate	O
results	O
and	O
user	O
studies	O
show	O
the	O
proposed	O
fusion	O
methods	O
are	O
practical	O
and	O
efficient	O
solutions	O
to	O
Web	B-KEY
object	I-KEY
ranking	B-KEY
in	O
cases	O
we	O
have	O
described	O
.	O
Though	O
the	O
experiments	O
were	O
conducted	O
on	O
high-quality	O
photo	O
ranking	B-KEY
,	O
the	O
proposed	O
algorithms	B-KEY
are	O
also	O
applicable	O
to	O
other	O
ranking	B-KEY
problems	O
,	O
such	O
as	O
movie	O
ranking	B-KEY
and	O
music	O
ranking	B-KEY
.	O

A	O
Dynamic	O
Pari-Mutuel	B-KEY
Market	I-KEY
for	O
Hedging	O
,	O
Wagering	O
,	O
and	O
Information	O
Aggregation	O
ABSTRACT	O
I	O
develop	O
a	O
new	O
mechanism	O
for	O
risk	B-KEY
allocation	I-KEY
and	O
information	B-KEY
speculation	I-KEY
called	O
a	O
dynamic	O
pari-mutuel	O
market	O
-LRB-	O
DPM	O
-RRB-	O
.	O
A	O
DPM	B-KEY
acts	O
as	O
hybrid	B-KEY
between	O
a	O
pari-mutuel	B-KEY
market	I-KEY
and	O
a	O
continuous	B-KEY
double	I-KEY
auction	I-KEY
-LRB-	O
CDA	B-KEY
-RRB-	O
,	O
inheriting	O
some	O
of	O
the	O
advantages	O
of	O
both	O
.	O
Like	O
a	O
pari-mutuel	B-KEY
market	I-KEY
,	O
a	O
DPM	B-KEY
offers	O
infinite	O
buy-in	O
liquidity	O
and	O
zero	B-KEY
risk	I-KEY
for	O
the	O
market	B-KEY
institution	I-KEY
;	O
like	O
a	O
CDA	B-KEY
,	O
a	O
DPM	B-KEY
can	O
continuously	O
react	O
to	O
new	O
information	O
,	O
dynamically	O
incorporate	O
information	O
into	O
prices	B-KEY
,	O
and	O
allow	O
traders	O
to	O
lock	O
in	O
gains	B-KEY
or	O
limit	O
losses	B-KEY
by	O
selling	B-KEY
prior	O
to	O
event	B-KEY
resolution	I-KEY
.	O
The	O
trader	B-KEY
interface	I-KEY
can	O
be	O
designed	O
to	O
mimic	O
the	O
familiar	O
double	B-KEY
auction	I-KEY
format	I-KEY
with	O
bid-ask	B-KEY
queues	I-KEY
,	O
though	O
with	O
an	O
addition	O
variable	O
called	O
the	O
payoff	B-KEY
per	I-KEY
share	I-KEY
.	O
The	O
DPM	B-KEY
price	B-KEY
function	O
can	O
be	O
viewed	O
as	O
an	O
automated	O
market	O
maker	O
always	O
offering	O
to	O
sell	O
at	O
some	O
price	O
,	O
and	O
moving	O
the	O
price	O
appropriately	O
according	O
to	O
demand	O
.	O
Since	O
the	O
mechanism	O
is	O
pari-mutuel	O
-LRB-	O
i.e.	O
,	O
redistributive	O
-RRB-	O
,	O
it	O
is	O
guaranteed	O
to	O
pay	O
out	O
exactly	O
the	O
amount	O
of	O
money	O
taken	O
in	O
.	O
I	O
explore	O
a	O
number	O
of	O
variations	O
on	O
the	O
basic	O
DPM	B-KEY
,	O
analyzing	O
the	O
properties	O
of	O
each	O
,	O
and	O
solving	O
in	O
closed	O
form	O
for	O
their	O
respective	O
price	B-KEY
functions	O
.	O

Playing	O
Games	O
in	O
Many	O
Possible	O
Worlds	O
ABSTRACT	O
In	O
traditional	O
game	B-KEY
theory	I-KEY
,	O
players	O
are	O
typically	O
endowed	O
with	O
exogenously	O
given	O
knowledge	O
of	O
the	O
structure	O
of	O
the	O
game	O
--	O
either	O
full	O
omniscient	O
knowledge	O
or	O
partial	O
but	O
fixed	O
information	O
.	O
In	O
real	O
life	O
,	O
however	O
,	O
people	O
are	O
often	O
unaware	O
of	O
the	O
utility	O
of	O
taking	O
a	O
particular	O
action	O
until	O
they	O
perform	O
research	O
into	O
its	O
consequences	O
.	O
In	O
this	O
paper	O
,	O
we	O
model	O
this	O
phenomenon	O
.	O
We	O
imagine	O
a	O
player	O
engaged	O
in	O
a	O
questionand-answer	O
session	O
,	O
asking	O
questions	O
both	O
about	O
his	O
or	O
her	O
own	O
preferences	O
and	O
about	O
the	O
state	O
of	O
reality	O
;	O
thus	O
we	O
call	O
this	O
setting	O
``	O
Socratic	O
''	O
game	B-KEY
theory	I-KEY
.	O
In	O
a	O
Socratic	B-KEY
game	I-KEY
,	O
players	O
begin	O
with	O
an	O
a	O
priori	B-KEY
probability	I-KEY
distribution	I-KEY
over	O
many	O
possible	O
worlds	O
,	O
with	O
a	O
different	O
utility	O
function	O
for	O
each	O
world	O
.	O
Players	O
can	O
make	O
queries	O
,	O
at	O
some	O
cost	O
,	O
to	O
learn	O
partial	O
information	O
about	O
which	O
of	O
the	O
possible	O
worlds	O
is	O
the	O
actual	O
world	O
,	O
before	O
choosing	O
an	O
action	O
.	O
We	O
consider	O
two	O
query	O
models	O
:	O
-LRB-	O
1	O
-RRB-	O
an	O
unobservable-query	O
model	O
,	O
in	O
which	O
players	O
learn	O
only	O
the	O
response	O
to	O
their	O
own	O
queries	O
,	O
and	O
-LRB-	O
2	O
-RRB-	O
an	O
observable-query	O
model	O
,	O
in	O
which	O
players	O
also	O
learn	O
which	O
queries	O
their	O
opponents	O
made	O
.	O
The	O
results	O
in	O
this	O
paper	O
consider	O
cases	O
in	O
which	O
the	O
underlying	O
worlds	O
of	O
a	O
two-player	O
Socratic	B-KEY
game	I-KEY
are	O
either	O
constant-sum	B-KEY
games	I-KEY
or	O
strategically	O
zero-sum	O
games	O
,	O
a	O
class	O
that	O
generalizes	O
constant-sum	B-KEY
games	I-KEY
to	O
include	O
all	O
games	O
in	O
which	O
the	O
sum	O
of	O
payoffs	O
depends	O
linearly	O
on	O
the	O
interaction	O
between	O
the	O
players	O
.	O
When	O
the	O
underlying	O
worlds	O
are	O
constant	O
sum	O
,	O
we	O
give	O
polynomial-time	O
algorithms	B-KEY
to	O
find	O
Nash	O
equilibria	O
in	O
both	O
the	O
observable	O
-	O
and	O
unobservable-query	O
models	O
.	O
When	O
the	O
worlds	O
are	O
strategically	O
zero	O
sum	O
,	O
we	O
give	O
efficient	O
algorithms	B-KEY
to	O
find	O
Nash	O
equilibria	O
in	O
unobservablequery	O
Socratic	B-KEY
games	I-KEY
and	O
correlated	O
equilibria	O
in	O
observablequery	O
Socratic	B-KEY
games	I-KEY
.	O

A	O
Study	O
of	O
Factors	O
Affecting	O
the	O
Utility	O
of	O
Implicit	B-KEY
Relevance	I-KEY
Feedback	I-KEY
ABSTRACT	O
Implicit	B-KEY
relevance	I-KEY
feedback	I-KEY
-LRB-	O
IRF	O
-RRB-	O
is	O
the	O
process	O
by	O
which	O
a	O
search	O
system	O
unobtrusively	O
gathers	O
evidence	O
on	O
searcher	O
interests	O
from	O
their	O
interaction	O
with	O
the	O
system	O
.	O
IRF	O
is	O
a	O
new	O
method	O
of	O
gathering	O
information	O
on	O
user	O
interest	O
and	O
,	O
if	O
IRF	O
is	O
to	O
be	O
used	O
in	O
operational	O
IR	O
systems	O
,	O
it	O
is	O
important	O
to	O
establish	O
when	O
it	O
performs	O
well	O
and	O
when	O
it	O
performs	O
poorly	O
.	O
In	O
this	O
paper	O
we	O
investigate	O
how	O
the	O
use	O
and	O
effectiveness	O
of	O
IRF	O
is	O
affected	O
by	O
three	O
factors	O
:	O
search	B-KEY
task	I-KEY
complexity	I-KEY
,	O
the	O
search	O
experience	O
of	O
the	O
user	O
and	O
the	O
stage	O
in	O
the	O
search	O
.	O
Our	O
findings	O
suggest	O
that	O
all	O
three	O
of	O
these	O
factors	O
contribute	O
to	O
the	O
utility	O
of	O
IRF	O
.	O

Multi-Attribute	O
Coalitional	B-KEY
Games	I-KEY
*	O
t	O
ABSTRACT	O
We	O
study	O
coalitional	B-KEY
games	I-KEY
where	O
the	O
value	O
of	O
cooperation	B-KEY
among	O
the	O
agents	B-KEY
are	O
solely	O
determined	O
by	O
the	O
attributes	O
the	O
agents	B-KEY
possess	O
,	O
with	O
no	O
assumption	O
as	O
to	O
how	O
these	O
attributes	O
jointly	O
determine	O
this	O
value	O
.	O
This	O
framework	O
allows	O
us	O
to	O
model	O
diverse	B-KEY
economic	I-KEY
interactions	I-KEY
by	O
picking	O
the	O
right	O
attributes	O
.	O
We	O
study	O
the	O
computational	B-KEY
complexity	I-KEY
of	O
two	O
coalitional	O
solution	O
concepts	O
for	O
these	O
games	O
--	O
the	O
Shapley	O
value	O
and	O
the	O
core	B-KEY
.	O
We	O
show	O
how	O
the	O
positive	O
results	O
obtained	O
in	O
this	O
paper	O
imply	O
comparable	O
results	O
for	O
other	O
games	O
studied	O
in	O
the	O
literature	O
.	O

Weak	B-KEY
Monotonicity	I-KEY
Suffices	O
for	O
Truthfulness	B-KEY
on	O
Convex	B-KEY
Domains	I-KEY
ABSTRACT	O
Weak	B-KEY
monotonicity	I-KEY
is	O
a	O
simple	O
necessary	O
condition	O
for	O
a	O
social	B-KEY
choice	I-KEY
function	I-KEY
to	O
be	O
implementable	O
by	O
a	O
truthful	B-KEY
mechanism	O
.	O
Roberts	O
-LSB-	O
10	O
-RSB-	O
showed	O
that	O
it	O
is	O
sufficient	O
for	O
all	O
social	B-KEY
choice	I-KEY
functions	I-KEY
whose	O
domain	O
is	O
unrestricted	O
.	O
Lavi	O
,	O
Mu'alem	O
and	O
Nisan	O
-LSB-	O
6	O
-RSB-	O
proved	O
the	O
sufficiency	O
of	O
weak	B-KEY
monotonicity	I-KEY
for	O
functions	O
over	O
order-based	O
domains	O
and	O
Gui	O
,	O
Muller	O
and	O
Vohra	O
-LSB-	O
5	O
-RSB-	O
proved	O
sufficiency	O
for	O
order-based	O
domains	O
with	O
range	O
constraints	O
and	O
for	O
domains	O
defined	O
by	O
other	O
special	O
types	O
of	O
linear	O
inequality	O
constraints	O
.	O
Here	O
we	O
show	O
the	O
more	O
general	O
result	O
,	O
conjectured	O
by	O
Lavi	O
,	O
Mu'alem	O
and	O
Nisan	O
-LSB-	O
6	O
-RSB-	O
,	O
that	O
weak	B-KEY
monotonicity	I-KEY
is	O
sufficient	O
for	O
functions	O
defined	O
on	O
any	O
convex	B-KEY
domain	I-KEY
.	O

Using	O
Asymmetric	O
Distributions	O
to	O
Improve	O
Text	B-KEY
Classifier	I-KEY
Probability	B-KEY
Estimates	I-KEY
ABSTRACT	O
Text	B-KEY
classifiers	I-KEY
that	O
give	O
probability	B-KEY
estimates	I-KEY
are	O
more	O
readily	O
applicable	O
in	O
a	O
variety	O
of	O
scenarios	O
.	O
For	O
example	O
,	O
rather	O
than	O
choosing	O
one	O
set	O
decision	B-KEY
threshold	I-KEY
,	O
they	O
can	O
be	O
used	O
in	O
a	O
Bayesian	B-KEY
risk	I-KEY
model	I-KEY
to	O
issue	O
a	O
run-time	O
decision	O
which	O
minimizes	O
a	O
userspecified	O
cost	O
function	O
dynamically	O
chosen	O
at	O
prediction	O
time	O
.	O
However	O
,	O
the	O
quality	O
of	O
the	O
probability	B-KEY
estimates	I-KEY
is	O
crucial	O
.	O
We	O
review	O
a	O
variety	O
of	O
standard	O
approaches	O
to	O
converting	O
scores	O
-LRB-	O
and	O
poor	O
probability	B-KEY
estimates	I-KEY
-RRB-	O
from	O
text	B-KEY
classifiers	I-KEY
to	O
high	O
quality	O
estimates	O
and	O
introduce	O
new	O
models	O
motivated	O
by	O
the	O
intuition	O
that	O
the	O
empirical	B-KEY
score	I-KEY
distribution	I-KEY
for	O
the	O
``	O
extremely	O
irrelevant	O
''	O
,	O
``	O
hard	O
to	O
discriminate	O
''	O
,	O
and	O
``	O
obviously	O
relevant	O
''	O
items	O
are	O
often	O
significantly	O
different	O
.	O
Finally	O
,	O
we	O
analyze	O
the	O
experimental	O
performance	O
of	O
these	O
models	O
over	O
the	O
outputs	O
of	O
two	O
text	B-KEY
classifiers	I-KEY
.	O
The	O
analysis	O
demonstrates	O
that	O
one	O
of	O
these	O
models	O
is	O
theoretically	O
attractive	O
-LRB-	O
introducing	O
few	O
new	O
parameters	O
while	O
increasing	O
flexibility	O
-RRB-	O
,	O
computationally	O
efficient	O
,	O
and	O
empirically	O
preferable	O
.	O

Adaptive	O
Duty	B-KEY
Cycling	I-KEY
for	O
Energy	B-KEY
Harvesting	I-KEY
Systems	O
-LCB-	O
jasonh	O
,	O
kansal	O
,	O
szahedi	O
,	O
mbs	O
-RCB-	O
@	O
ee.ucla.edu	O
vijay@nec-labs.com	O
ABSTRACT	O
Harvesting	O
energy	O
from	O
the	O
environment	O
is	O
feasible	O
in	O
many	O
applications	O
to	O
ameliorate	O
the	O
energy	O
limitations	O
in	O
sensor	B-KEY
networks	I-KEY
.	O
In	O
this	O
paper	O
,	O
we	O
present	O
an	O
adaptive	O
duty	B-KEY
cycling	I-KEY
algorithm	O
that	O
allows	O
energy	B-KEY
harvesting	I-KEY
sensor	O
nodes	O
to	O
autonomously	O
adjust	O
their	O
duty	B-KEY
cycle	I-KEY
according	O
to	O
the	O
energy	O
availability	O
in	O
the	O
environment	O
.	O
The	O
algorithm	O
has	O
three	O
objectives	O
,	O
namely	O
-LRB-	O
a	O
-RRB-	O
achieving	O
energy	B-KEY
neutral	I-KEY
operation	I-KEY
,	O
i.e.	O
,	O
energy	O
consumption	O
should	O
not	O
be	O
more	O
than	O
the	O
energy	O
provided	O
by	O
the	O
environment	O
,	O
-LRB-	O
b	O
-RRB-	O
maximizing	O
the	O
system	O
performance	O
based	O
on	O
an	O
application	O
utility	O
model	O
subject	O
to	O
the	O
above	O
energyneutrality	O
constraint	O
,	O
and	O
-LRB-	O
c	O
-RRB-	O
adapting	O
to	O
the	O
dynamics	O
of	O
the	O
energy	O
source	O
at	O
run-time	O
.	O
We	O
present	O
a	O
model	O
that	O
enables	O
harvesting	O
sensor	O
nodes	O
to	O
predict	O
future	O
energy	O
opportunities	O
based	O
on	O
historical	O
data	O
.	O
We	O
also	O
derive	O
an	O
upper	O
bound	O
on	O
the	O
maximum	O
achievable	O
performance	O
assuming	O
perfect	O
knowledge	O
about	O
the	O
future	O
behavior	O
of	O
the	O
energy	O
source	O
.	O
Our	O
methods	O
are	O
evaluated	O
using	O
data	O
gathered	O
from	O
a	O
prototype	O
solar	O
energy	B-KEY
harvesting	I-KEY
platform	O
and	O
we	O
show	O
that	O
our	O
algorithm	O
can	O
utilize	O
up	O
to	O
58	O
%	O
more	O
environmental	B-KEY
energy	I-KEY
compared	O
to	O
the	O
case	O
when	O
harvesting-aware	O
power	B-KEY
management	I-KEY
is	O
not	O
used	O
.	O

A	O
Complete	O
Distributed	B-KEY
Constraint	I-KEY
Optimization	I-KEY
Method	O
For	O
Non-Traditional	O
Pseudotree	B-KEY
Arrangements	I-KEY
*	O
ABSTRACT	O
Distributed	B-KEY
Constraint	I-KEY
Optimization	I-KEY
-LRB-	O
DCOP	O
-RRB-	O
is	O
a	O
general	O
framework	O
that	O
can	O
model	O
complex	O
problems	O
in	O
multi-agent	O
systems	O
.	O
Several	O
current	O
algorithms	O
that	O
solve	O
general	O
DCOP	O
instances	O
,	O
including	O
ADOPT	O
and	O
DPOP	O
,	O
arrange	O
agents	B-KEY
into	O
a	O
traditional	O
pseudotree	O
structure	O
.	O
We	O
introduce	O
an	O
extension	O
to	O
the	O
DPOP	O
algorithm	O
that	O
handles	O
an	O
extended	O
set	O
of	O
pseudotree	B-KEY
arrangements	I-KEY
.	O
Our	O
algorithm	O
correctly	O
solves	O
DCOP	O
instances	O
for	O
pseudotrees	O
that	O
include	O
edges	O
between	O
nodes	O
in	O
separate	O
branches	O
.	O
The	O
algorithm	O
also	O
solves	O
instances	O
with	O
traditional	O
pseudotree	B-KEY
arrangements	I-KEY
using	O
the	O
same	O
procedure	O
as	O
DPOP	O
.	O
We	O
compare	O
our	O
algorithm	O
with	O
DPOP	O
using	O
several	O
metrics	O
including	O
the	O
induced	O
width	O
of	O
the	O
pseudotrees	O
,	O
the	O
maximum	O
dimensionality	O
of	O
messages	O
and	O
computation	O
,	O
and	O
the	O
maximum	B-KEY
sequential	I-KEY
path	I-KEY
cost	I-KEY
through	O
the	O
algorithm	O
.	O
We	O
prove	O
that	O
for	O
some	O
problem	O
instances	O
it	O
is	O
not	O
possible	O
to	O
generate	O
a	O
traditional	O
pseudotree	O
using	O
edge-traversal	O
heuristics	O
that	O
will	O
outperform	O
a	O
cross-edged	B-KEY
pseudotree	I-KEY
.	O
We	O
use	O
multiple	O
heuristics	O
to	O
generate	O
pseudotrees	O
and	O
choose	O
the	O
best	O
pseudotree	O
in	O
linear	O
space-time	O
complexity	O
.	O
For	O
some	O
problem	O
instances	O
we	O
observe	O
significant	O
improvements	O
in	O
message	O
and	O
computation	O
sizes	O
compared	O
to	O
DPOP	O
.	O

Hypotheses	O
Refinement	O
under	O
Topological	O
Communication	O
Constraints	O
*	O
ABSTRACT	O
We	O
investigate	O
the	O
properties	O
of	O
a	O
multiagent	B-KEY
system	I-KEY
where	O
each	O
-LRB-	O
distributed	O
-RRB-	O
agent	O
locally	O
perceives	O
its	O
environment	O
.	O
Upon	O
perception	O
of	O
an	O
unexpected	O
event	O
,	O
each	O
agent	O
locally	O
computes	O
its	O
favoured	B-KEY
hypothesis	I-KEY
and	O
tries	O
to	O
propagate	O
it	O
to	O
other	O
agents	O
,	O
by	O
exchanging	O
hypotheses	O
and	O
supporting	O
arguments	O
-LRB-	O
observations	O
-RRB-	O
.	O
However	O
,	O
we	O
further	O
assume	O
that	O
communication	O
opportunities	O
are	O
severely	O
constrained	O
and	O
change	O
dynamically	O
.	O
In	O
this	O
paper	O
,	O
we	O
mostly	O
investigate	O
the	O
convergence	O
of	O
such	O
systems	O
towards	O
global	B-KEY
consistency	I-KEY
.	O
We	O
first	O
show	O
that	O
-LRB-	O
for	O
a	O
wide	O
class	O
of	O
protocols	O
that	O
we	O
shall	O
define	O
-RRB-	O
,	O
the	O
communication	O
constraints	O
induced	O
by	O
the	O
topology	O
will	O
not	O
prevent	O
the	O
convergence	O
of	O
the	O
system	O
,	O
at	O
the	O
condition	O
that	O
the	O
system	O
dynamics	O
guarantees	O
that	O
no	O
agent	O
will	O
ever	O
be	O
isolated	O
forever	O
,	O
and	O
that	O
agents	O
have	O
unlimited	O
time	O
for	O
computation	O
and	O
arguments	O
exchange	O
.	O
As	O
this	O
assumption	O
can	O
not	O
be	O
made	O
in	O
most	O
situations	O
though	O
,	O
we	O
then	O
set	O
up	O
an	O
experimental	O
framework	O
aiming	O
at	O
comparing	O
the	O
relative	O
efficiency	O
and	O
effectiveness	O
of	O
different	O
interaction	O
protocols	O
for	O
hypotheses	O
exchange	O
.	O
We	O
study	O
a	O
critical	O
situation	O
involving	O
a	O
number	O
of	O
agents	O
aiming	O
at	O
escaping	O
from	O
a	O
burning	O
building	O
.	O
The	O
results	O
reported	O
here	O
provide	O
some	O
insights	O
regarding	O
the	O
design	O
of	O
optimal	O
protocol	O
for	O
hypotheses	O
refinement	O
in	O
this	O
context	O
.	O

Cross-Lingual	O
Query	B-KEY
Suggestion	I-KEY
Using	O
Query	B-KEY
Logs	I-KEY
of	O
Different	O
Languages	O
ABSTRACT	O
Query	B-KEY
suggestion	I-KEY
aims	O
to	O
suggest	O
relevant	O
queries	O
for	O
a	O
given	O
query	O
,	O
which	O
help	O
users	O
better	O
specify	O
their	O
information	O
needs	O
.	O
Previously	O
,	O
the	O
suggested	O
terms	O
are	O
mostly	O
in	O
the	O
same	O
language	O
of	O
the	O
input	O
query	O
.	O
In	O
this	O
paper	O
,	O
we	O
extend	O
it	O
to	O
cross-lingual	O
query	B-KEY
suggestion	I-KEY
-LRB-	O
CLQS	O
-RRB-	O
:	O
for	O
a	O
query	O
in	O
one	O
language	O
,	O
we	O
suggest	O
similar	O
or	O
relevant	O
queries	O
in	O
other	O
languages	O
.	O
This	O
is	O
very	O
important	O
to	O
scenarios	O
of	O
cross-language	B-KEY
information	I-KEY
retrieval	I-KEY
-LRB-	O
CLIR	O
-RRB-	O
and	O
cross-lingual	O
keyword	B-KEY
bidding	I-KEY
for	O
search	B-KEY
engine	I-KEY
advertisement	O
.	O
Instead	O
of	O
relying	O
on	O
existing	O
query	B-KEY
translation	I-KEY
technologies	O
for	O
CLQS	O
,	O
we	O
present	O
an	O
effective	O
means	O
to	O
map	B-KEY
the	O
input	O
query	O
of	O
one	O
language	O
to	O
queries	O
of	O
the	O
other	O
language	O
in	O
the	O
query	B-KEY
log	I-KEY
.	O
Important	O
monolingual	O
and	O
cross-lingual	O
information	O
such	O
as	O
word	O
translation	O
relations	O
and	O
word	O
co-occurrence	O
statistics	O
,	O
etc.	O
are	O
used	O
to	O
estimate	O
the	O
cross-lingual	O
query	O
similarity	O
with	O
a	O
discriminative	O
model	O
.	O
Benchmarks	B-KEY
show	O
that	O
the	O
resulting	O
CLQS	O
system	O
significantly	O
outperforms	O
a	O
baseline	O
system	O
based	O
on	O
dictionary-based	O
query	B-KEY
translation	I-KEY
.	O
Besides	O
,	O
the	O
resulting	O
CLQS	O
is	O
tested	O
with	O
French	O
to	O
English	O
CLIR	O
tasks	O
on	O
TREC	O
collections	O
.	O
The	O
results	O
demonstrate	O
higher	O
effectiveness	O
than	O
the	O
traditional	O
query	B-KEY
translation	I-KEY
methods	O
.	O

Networks	B-KEY
Preserving	O
Evolutionary	O
Equilibria	O
and	O
the	O
Power	B-KEY
of	I-KEY
Randomization	I-KEY
We	O
study	O
a	O
natural	O
extension	O
of	O
classical	O
evolutionary	B-KEY
game	I-KEY
theory	I-KEY
to	O
a	O
setting	O
in	O
which	O
pairwise	O
interactions	O
are	O
restricted	O
to	O
the	O
edges	O
of	O
an	O
undirected	O
graph	O
or	O
network	O
.	O
We	O
generalize	O
the	O
definition	O
of	O
an	O
evolutionary	B-KEY
stable	I-KEY
strategy	I-KEY
-LRB-	O
ESS	O
-RRB-	O
,	O
and	O
show	O
a	O
pair	O
of	O
complementary	O
results	O
that	O
exhibit	O
the	O
power	B-KEY
of	I-KEY
randomization	I-KEY
in	O
our	O
setting	O
:	O
subject	O
to	O
degree	O
or	O
edge	B-KEY
density	I-KEY
conditions	I-KEY
,	O
the	O
classical	O
ESS	O
of	O
any	O
game	O
are	O
preserved	O
when	O
the	O
graph	O
is	O
chosen	O
randomly	O
and	O
the	O
mutation	B-KEY
set	I-KEY
is	O
chosen	O
adversarially	O
,	O
or	O
when	O
the	O
graph	O
is	O
chosen	O
adversarially	O
and	O
the	O
mutation	B-KEY
set	I-KEY
is	O
chosen	O
randomly	O
.	O
We	O
examine	O
natural	B-KEY
strengthenings	I-KEY
of	O
our	O
generalized	O
ESS	O
definition	O
,	O
and	O
show	O
that	O
similarly	O
strong	O
results	O
are	O
not	O
possible	O
for	O
them	O
.	O

Bid	B-KEY
Expressiveness	O
and	O
Clearing	O
Algorithms	O
in	O
Multiattribute	O
Double	O
Auctions	B-KEY
ABSTRACT	O
We	O
investigate	O
the	O
space	O
of	O
two-sided	O
multiattribute	B-KEY
auctions	I-KEY
,	O
focusing	O
on	O
the	O
relationship	O
between	O
constraints	O
on	O
the	O
offers	O
traders	O
can	O
express	O
through	O
bids	O
,	O
and	O
the	O
resulting	O
computational	O
problem	O
of	O
determining	O
an	O
optimal	O
set	O
of	O
trades	O
.	O
We	O
develop	O
a	O
formal	O
semantic	B-KEY
framework	I-KEY
for	O
characterizing	O
expressible	O
offers	O
,	O
and	O
show	O
conditions	O
under	O
which	O
the	O
allocation	O
problem	O
can	O
be	O
separated	O
into	O
first	O
identifying	O
optimal	O
pairwise	O
trades	O
and	O
subsequently	O
optimizing	O
combinations	O
of	O
those	O
trades	O
.	O
We	O
analyze	O
the	O
bilateral	O
matching	O
problem	O
while	O
taking	O
into	O
consideration	O
relevant	O
results	O
from	O
multiattribute	B-KEY
utility	I-KEY
theory	I-KEY
.	O
Network	O
flow	O
models	O
we	O
develop	O
for	O
computing	O
global	B-KEY
allocations	I-KEY
facilitate	O
classification	O
of	O
the	O
problem	O
space	O
by	O
computational	O
complexity	O
,	O
and	O
provide	O
guidance	O
for	O
developing	O
solution	O
algorithms	O
.	O
Experimental	O
trials	O
help	O
distinguish	O
tractable	O
problem	O
classes	O
for	O
proposed	O
solution	O
techniques	O
.	O

Privacy	B-KEY
in	O
Electronic	B-KEY
Commerce	I-KEY
and	O
the	O
Economics	O
of	O
Immediate	B-KEY
Gratification	I-KEY
ABSTRACT	O
Dichotomies	O
between	O
privacy	B-KEY
attitudes	O
and	O
behavior	O
have	O
been	O
noted	O
in	O
the	O
literature	O
but	O
not	O
yet	O
fully	O
explained	O
.	O
We	O
apply	O
lessons	O
from	O
the	O
research	O
on	O
behavioral	O
economics	O
to	O
understand	O
the	O
individual	B-KEY
decision	I-KEY
making	I-KEY
process	I-KEY
with	O
respect	O
to	O
privacy	B-KEY
in	O
electronic	B-KEY
commerce	I-KEY
.	O
We	O
show	O
that	O
it	O
is	O
unrealistic	O
to	O
expect	O
individual	O
rationality	B-KEY
in	O
this	O
context	O
.	O
Models	O
of	O
self-control	B-KEY
problems	I-KEY
and	O
immediate	B-KEY
gratification	I-KEY
offer	O
more	O
realistic	O
descriptions	O
of	O
the	O
decision	O
process	O
and	O
are	O
more	O
consistent	O
with	O
currently	O
available	O
data	O
.	O
In	O
particular	O
,	O
we	O
show	O
why	O
individuals	O
who	O
may	O
genuinely	O
want	O
to	O
protect	O
their	O
privacy	B-KEY
might	O
not	O
do	O
so	O
because	O
of	O
psychological	B-KEY
distortions	I-KEY
well	O
documented	O
in	O
the	O
behavioral	O
literature	O
;	O
we	O
show	O
that	O
these	O
distortions	O
may	O
affect	O
not	O
only	O
`	O
na	O
¨	O
ıve	O
'	O
individuals	O
but	O
also	O
`	O
sophisticated	O
'	O
ones	O
;	O
and	O
we	O
prove	O
that	O
this	O
may	O
occur	O
also	O
when	O
individuals	O
perceive	O
the	O
risks	O
from	O
not	O
protecting	O
their	O
privacy	B-KEY
as	O
significant	O
.	O

Bidding	B-KEY
Algorithms	I-KEY
for	O
a	O
Distributed	O
Combinatorial	B-KEY
Auction	I-KEY
ABSTRACT	O
Distributed	B-KEY
allocation	I-KEY
and	O
multiagent	O
coordination	B-KEY
problems	O
can	O
be	O
solved	O
through	O
combinatorial	B-KEY
auctions	I-KEY
.	O
However	O
,	O
most	O
of	O
the	O
existing	O
winner	O
determination	O
algorithms	O
for	O
combinatorial	B-KEY
auctions	I-KEY
are	O
centralized	O
.	O
The	O
PAUSE	B-KEY
auction	I-KEY
is	O
one	O
of	O
a	O
few	O
efforts	O
to	O
release	O
the	O
auctioneer	O
from	O
having	O
to	O
do	O
all	O
the	O
work	O
-LRB-	O
it	O
might	O
even	O
be	O
possible	O
to	O
get	O
rid	O
of	O
the	O
auctioneer	O
-RRB-	O
.	O
It	O
is	O
an	O
increasing	O
price	O
combinatorial	B-KEY
auction	I-KEY
that	O
naturally	O
distributes	O
the	O
problem	O
of	O
winner	O
determination	O
amongst	O
the	O
bidders	O
in	O
such	O
a	O
way	O
that	O
they	O
have	O
an	O
incentive	O
to	O
perform	O
the	O
calculation	O
.	O
It	O
can	O
be	O
used	O
when	O
we	O
wish	O
to	O
distribute	O
the	O
computational	O
load	O
among	O
the	O
bidders	O
or	O
when	O
the	O
bidders	O
do	O
not	O
wish	O
to	O
reveal	O
their	O
true	O
valuations	O
unless	O
necessary	O
.	O
PAUSE	O
establishes	O
the	O
rules	O
the	O
bidders	O
must	O
obey	O
.	O
However	O
,	O
it	O
does	O
not	O
tell	O
us	O
how	O
the	O
bidders	O
should	O
calculate	O
their	O
bids	O
.	O
We	O
have	O
developed	O
a	O
couple	O
of	O
bidding	B-KEY
algorithms	I-KEY
for	O
the	O
bidders	O
in	O
a	O
PAUSE	B-KEY
auction	I-KEY
.	O
Our	O
algorithms	O
always	O
return	O
the	O
set	O
of	O
bids	O
that	O
maximizes	O
the	O
bidder	O
's	O
utility	O
.	O
Since	O
the	O
problem	O
is	O
NP-Hard	O
,	O
run	O
time	O
remains	O
exponential	O
on	O
the	O
number	O
of	O
items	O
,	O
but	O
it	O
is	O
remarkably	O
better	O
than	O
an	O
exhaustive	O
search	O
.	O
In	O
this	O
paper	O
we	O
present	O
our	O
bidding	B-KEY
algorithms	I-KEY
,	O
discuss	O
their	O
virtues	O
and	O
drawbacks	O
,	O
and	O
compare	O
the	O
solutions	O
obtained	O
by	O
them	O
to	O
the	O
revenue-maximizing	O
solution	O
found	O
by	O
a	O
centralized	O
winner	O
determination	O
algorithm	O
.	O

Unifying	O
Distributed	O
Constraint	B-KEY
Algorithms	B-KEY
in	O
a	O
BDI	B-KEY
Negotiation	B-KEY
Framework	O
ABSTRACT	O
This	O
paper	O
presents	O
a	O
novel	O
,	O
unified	O
distributed	O
constraint	B-KEY
satisfaction	O
framework	O
based	O
on	O
automated	O
negotiation	B-KEY
.	O
The	O
Distributed	O
Constraint	B-KEY
Satisfaction	O
Problem	O
-LRB-	O
DCSP	O
-RRB-	O
is	O
one	O
that	O
entails	O
several	O
agents	O
to	O
search	O
for	O
an	O
agreement	O
,	O
which	O
is	O
a	O
consistent	O
combination	O
of	O
actions	O
that	O
satisfies	O
their	O
mutual	O
constraints	O
in	O
a	O
shared	O
environment	O
.	O
By	O
anchoring	O
the	O
DCSP	B-KEY
search	O
on	O
automated	O
negotiation	B-KEY
,	O
we	O
show	O
that	O
several	O
well-known	O
DCSP	B-KEY
algorithms	B-KEY
are	O
actually	O
mechanisms	O
that	O
can	O
reach	O
agreements	O
through	O
a	O
common	O
Belief-Desire-Intention	O
-LRB-	O
BDI	B-KEY
-RRB-	O
protocol	O
,	O
but	O
using	O
different	O
strategies	O
.	O
A	O
major	O
motivation	O
for	O
this	O
BDI	B-KEY
framework	O
is	O
that	O
it	O
not	O
only	O
provides	O
a	O
conceptually	O
clearer	O
understanding	O
of	O
existing	O
DCSP	B-KEY
algorithms	B-KEY
from	O
an	O
agent	O
model	O
perspective	O
,	O
but	O
also	O
opens	O
up	O
the	O
opportunities	O
to	O
extend	O
and	O
develop	O
new	O
strategies	O
for	O
DCSP	B-KEY
.	O
To	O
this	O
end	O
,	O
a	O
new	O
strategy	O
called	O
Unsolicited	O
Mutual	O
Advice	O
-LRB-	O
UMA	B-KEY
-RRB-	O
is	O
proposed	O
.	O
Performance	O
evaluation	O
shows	O
that	O
the	O
UMA	B-KEY
strategy	O
can	O
outperform	O
some	O
existing	O
mechanisms	O
in	O
terms	O
of	O
computational	O
cycles	O
.	O

Assured	O
Service	O
Quality	O
by	O
Improved	O
Fault	B-KEY
Management	I-KEY
Service-Oriented	O
Event	B-KEY
Correlation	I-KEY
ABSTRACT	O
The	O
paradigm	O
shift	O
from	O
device-oriented	O
to	O
service-oriented	O
management	O
has	O
also	O
implications	O
to	O
the	O
area	O
of	O
event	B-KEY
correlation	I-KEY
.	O
Today	O
's	O
event	B-KEY
correlation	I-KEY
mainly	O
addresses	O
the	O
correlation	O
of	O
events	O
as	O
reported	O
from	O
management	O
tools	O
.	O
However	O
,	O
a	O
correlation	O
of	O
user	O
trouble	O
reports	O
concerning	O
services	O
should	O
also	O
be	O
performed	O
.	O
This	O
is	O
necessary	O
to	O
improve	O
the	O
resolution	O
time	O
and	O
to	O
reduce	O
the	O
effort	O
for	O
keeping	O
the	O
service	O
agreements	O
.	O
We	O
refer	O
to	O
such	O
a	O
type	O
of	O
correlation	O
as	O
service-oriented	O
event	B-KEY
correlation	I-KEY
.	O
The	O
necessity	O
to	O
use	O
this	O
kind	O
of	O
event	B-KEY
correlation	I-KEY
is	O
motivated	O
in	O
the	O
paper	O
.	O
To	O
introduce	O
service-oriented	O
event	B-KEY
correlation	I-KEY
for	O
an	O
IT	O
service	O
provider	O
,	O
an	O
appropriate	O
modeling	O
of	O
the	O
correlation	O
workflow	O
and	O
of	O
the	O
information	O
is	O
necessary	O
.	O
Therefore	O
,	O
we	O
examine	O
the	O
process	B-KEY
management	I-KEY
frameworks	I-KEY
IT	O
Infrastructure	O
Library	O
-LRB-	O
ITIL	O
-RRB-	O
and	O
enhanced	O
Telecom	O
Operations	O
Map	O
-LRB-	O
eTOM	O
-RRB-	O
for	O
their	O
contribution	O
to	O
the	O
workflow	O
modeling	O
in	O
this	O
area	O
.	O
The	O
different	O
kinds	O
of	O
dependencies	O
that	O
we	O
find	O
in	O
our	O
general	O
scenario	O
are	O
then	O
used	O
to	O
develop	O
a	O
workflow	O
for	O
the	O
service-oriented	O
event	B-KEY
correlation	I-KEY
.	O
The	O
MNM	O
Service	O
Model	O
,	O
which	O
is	O
a	O
generic	O
model	O
for	O
IT	O
service	B-KEY
management	I-KEY
proposed	O
by	O
the	O
Munich	O
Network	O
Management	O
-LRB-	O
MNM	O
-RRB-	O
Team	O
,	O
is	O
used	O
to	O
derive	O
an	O
appropriate	O
information	O
modeling	O
.	O
An	O
example	O
scenario	O
,	O
the	O
Web	O
Hosting	O
Service	O
of	O
the	O
Leibniz	O
Supercomputing	O
Center	O
-LRB-	O
LRZ	O
-RRB-	O
,	O
is	O
used	O
to	O
demonstrate	O
the	O
application	O
of	O
service-oriented	O
event	B-KEY
correlation	I-KEY
.	O

Applying	O
Learning	B-KEY
Algorithms	O
to	O
Preference	O
Elicitation	O
ABSTRACT	O
We	O
consider	O
the	O
parallels	B-KEY
between	O
the	O
preference	B-KEY
elicitation	I-KEY
problem	O
in	O
combinatorial	O
auctions	O
and	O
the	O
problem	O
of	O
learning	O
an	O
unknown	O
function	O
from	O
learning	O
theory	O
.	O
We	O
show	O
that	O
learning	B-KEY
algorithms	O
can	O
be	O
used	O
as	O
a	O
basis	O
for	O
preference	O
elicitation	O
algorithms	O
.	O
The	O
resulting	O
elicitation	B-KEY
algorithms	I-KEY
perform	O
a	O
polynomial	B-KEY
number	O
of	O
queries	O
.	O
We	O
also	O
give	O
conditions	O
under	O
which	O
the	O
resulting	B-KEY
algorithms	I-KEY
have	O
polynomial	B-KEY
communication	O
.	O
Our	O
conversion	B-KEY
procedure	I-KEY
allows	O
us	O
to	O
generate	O
combinatorial	B-KEY
auction	I-KEY
protocols	O
from	O
learning	O
algorithms	O
for	O
polynomials	O
,	O
monotone	O
DNF	O
,	O
and	O
linear-threshold	O
functions	O
.	O
In	O
particular	O
,	O
we	O
obtain	O
an	O
algorithm	O
that	O
elicits	O
XOR	B-KEY
bids	I-KEY
with	O
polynomial	B-KEY
communication	O
.	O

Query	B-KEY
Performance	I-KEY
Prediction	I-KEY
in	O
Web	B-KEY
Search	I-KEY
Environments	O
ABSTRACT	O
Current	O
prediction	O
techniques	O
,	O
which	O
are	O
generally	O
designed	O
for	O
content-based	O
queries	O
and	O
are	O
typically	O
evaluated	O
on	O
relatively	O
homogenous	B-KEY
test	I-KEY
collections	I-KEY
of	O
small	O
sizes	O
,	O
face	O
serious	O
challenges	O
in	O
web	B-KEY
search	I-KEY
environments	O
where	O
collections	O
are	O
significantly	O
more	O
heterogeneous	O
and	O
different	O
types	O
of	O
retrieval	O
tasks	O
exist	O
.	O
In	O
this	O
paper	O
,	O
we	O
present	O
three	O
techniques	O
to	O
address	O
these	O
challenges	O
.	O
We	O
focus	O
on	O
performance	O
prediction	O
for	O
two	O
types	O
of	O
queries	O
in	O
web	B-KEY
search	I-KEY
environments	O
:	O
content-based	O
and	O
Named-Page	O
finding	O
.	O
Our	O
evaluation	O
is	O
mainly	O
performed	O
on	O
the	O
GOV2	B-KEY
collection	I-KEY
.	O
In	O
addition	O
to	O
evaluating	O
our	O
models	O
for	O
the	O
two	O
types	O
of	O
queries	O
separately	O
,	O
we	O
consider	O
a	O
more	O
challenging	O
and	O
realistic	O
situation	O
that	O
the	O
two	O
types	O
of	O
queries	O
are	O
mixed	O
together	O
without	O
prior	O
information	O
on	O
query	O
types	O
.	O
To	O
assist	O
prediction	O
under	O
the	O
mixed-query	O
situation	O
,	O
a	O
novel	O
query	O
classifier	O
is	O
adopted	O
.	O
Results	O
show	O
that	O
our	O
prediction	O
of	O
web	O
query	O
performance	O
is	O
substantially	O
more	O
accurate	O
than	O
the	O
current	O
stateof-the-art	O
prediction	O
techniques	O
.	O
Consequently	O
,	O
our	O
paper	O
provides	O
a	O
practical	O
approach	O
to	O
performance	O
prediction	O
in	O
realworld	O
web	O
settings	O
.	O

A	O
Semantic	B-KEY
Approach	O
to	O
Contextual	B-KEY
Advertising	I-KEY
ABSTRACT	O
Contextual	B-KEY
advertising	I-KEY
or	O
Context	O
Match	B-KEY
-LRB-	O
CM	O
-RRB-	O
refers	O
to	O
the	O
placement	O
of	O
commercial	O
textual	O
advertisements	O
within	O
the	O
content	O
of	O
a	O
generic	O
web	O
page	O
,	O
while	O
Sponsored	O
Search	O
-LRB-	O
SS	O
-RRB-	O
advertising	O
consists	O
in	O
placing	O
ads	O
on	O
result	O
pages	O
from	O
a	O
web	O
search	O
engine	O
,	O
with	O
ads	O
driven	O
by	O
the	O
originating	O
query	O
.	O
In	O
CM	O
there	O
is	O
usually	O
an	O
intermediary	O
commercial	O
ad-network	O
entity	O
in	O
charge	O
of	O
optimizing	O
the	O
ad	O
selection	O
with	O
the	O
twin	O
goal	O
of	O
increasing	O
revenue	O
-LRB-	O
shared	O
between	O
the	O
publisher	O
and	O
the	O
ad-network	O
-RRB-	O
and	O
improving	O
the	O
user	O
experience	O
.	O
With	O
these	O
goals	O
in	O
mind	O
it	O
is	O
preferable	O
to	O
have	O
ads	B-KEY
relevant	I-KEY
to	O
the	O
page	O
content	O
,	O
rather	O
than	O
generic	O
ads	O
.	O
The	O
SS	O
market	O
developed	O
quicker	O
than	O
the	O
CM	O
market	O
,	O
and	O
most	O
textual	O
ads	O
are	O
still	O
characterized	O
by	O
``	O
bid	O
phrases	O
''	O
representing	O
those	O
queries	O
where	O
the	O
advertisers	O
would	O
like	O
to	O
have	O
their	O
ad	O
displayed	O
.	O
Hence	O
,	O
the	O
first	O
technologies	O
for	O
CM	O
have	O
relied	O
on	O
previous	O
solutions	O
for	O
SS	O
,	O
by	O
simply	O
extracting	O
one	O
or	O
more	O
phrases	O
from	O
the	O
given	O
page	O
content	O
,	O
and	O
displaying	O
ads	O
corresponding	O
to	O
searches	O
on	O
these	O
phrases	O
,	O
in	O
a	O
purely	O
syntactic	O
approach	O
.	O
However	O
,	O
due	O
to	O
the	O
vagaries	O
of	O
phrase	O
extraction	O
,	O
and	O
the	O
lack	O
of	O
context	O
,	O
this	O
approach	O
leads	O
to	O
many	O
irrelevant	O
ads	O
.	O
To	O
overcome	O
this	O
problem	O
,	O
we	O
propose	O
a	O
system	O
for	O
contextual	O
ad	O
matching	B-KEY
based	O
on	O
a	O
combination	O
of	O
semantic	B-KEY
and	O
syntactic	O
features	O
.	O

ICE	O
:	O
An	O
Iterative	B-KEY
Combinatorial	I-KEY
Exchange	I-KEY
David	O
C.	O
Parkes	O
*	O
s	O
Ruggiero	O
Cavallos	O
Nick	O
Elprins	O
Adam	O
Judas	O
S	O
´	O
ebastien	O
Lahaies	O
ABSTRACT	O
We	O
present	O
the	O
first	O
design	O
for	O
an	O
iterative	B-KEY
combinatorial	I-KEY
exchange	I-KEY
-LRB-	O
ICE	O
-RRB-	O
.	O
The	O
exchange	O
incorporates	O
a	O
tree-based	O
bidding	B-KEY
language	O
that	O
is	O
concise	O
and	O
expressive	O
for	O
CEs	O
.	O
Bidders	O
specify	O
lower	O
and	O
upper	O
bounds	O
on	O
their	O
value	O
for	O
different	O
trades	B-KEY
.	O
These	O
bounds	O
allow	O
price	B-KEY
discovery	O
and	O
useful	O
preference	B-KEY
elicitation	I-KEY
in	O
early	O
rounds	O
,	O
and	O
allow	O
termination	O
with	O
an	O
efficient	O
trade	B-KEY
despite	O
partial	O
information	O
on	O
bidder	O
valuations	O
.	O
All	O
computation	O
in	O
the	O
exchange	O
is	O
carefully	O
optimized	O
to	O
exploit	O
the	O
structure	O
of	O
the	O
bid-trees	O
and	O
to	O
avoid	O
enumerating	O
trades	O
.	O
A	O
proxied	O
interpretation	O
of	O
a	O
revealedpreference	O
activity	O
rule	O
ensures	O
progress	O
across	O
rounds	O
.	O
A	O
VCG-based	O
payment	O
scheme	O
that	O
has	O
been	O
shown	O
to	O
mitigate	O
opportunities	O
for	O
bargaining	O
and	O
strategic	O
behavior	O
is	O
used	O
to	O
determine	O
final	O
payments	O
.	O
The	O
exchange	O
is	O
fully	O
implemented	O
and	O
in	O
a	O
validation	O
phase	O
.	O

Normative	B-KEY
System	I-KEY
Games	O
ABSTRACT	O
We	O
develop	O
a	O
model	O
of	O
normative	B-KEY
systems	I-KEY
in	O
which	O
agents	O
are	O
assumed	O
to	O
have	O
multiple	O
goals	B-KEY
of	O
increasing	O
priority	O
,	O
and	O
investigate	O
the	O
computational	O
complexity	O
and	O
game	O
theoretic	O
properties	O
of	O
this	O
model	O
.	O
In	O
the	O
underlying	O
model	O
of	O
normative	B-KEY
systems	I-KEY
,	O
we	O
use	O
Kripke	B-KEY
structures	I-KEY
to	O
represent	O
the	O
possible	O
transitions	O
of	O
a	O
multiagent	O
system	O
.	O
A	O
normative	B-KEY
system	I-KEY
is	O
then	O
simply	O
a	O
subset	O
of	O
the	O
Kripke	B-KEY
structure	I-KEY
,	O
which	O
contains	O
the	O
arcs	O
that	O
are	O
forbidden	O
by	O
the	O
normative	B-KEY
system	I-KEY
.	O
We	O
specify	O
an	O
agent	O
's	O
goals	B-KEY
as	O
a	O
hierarchy	O
of	O
formulae	O
of	O
Computation	B-KEY
Tree	I-KEY
Logic	I-KEY
-LRB-	O
CTL	O
-RRB-	O
,	O
a	O
widely	O
used	O
logic	O
for	O
representing	O
the	O
properties	O
of	O
Kripke	O
structures	O
:	O
the	O
intuition	O
is	O
that	O
goals	O
further	O
up	O
the	O
hierarchy	O
are	O
preferred	O
by	O
the	O
agent	O
over	O
those	O
that	O
appear	O
further	O
down	O
the	O
hierarchy	O
.	O
Using	O
this	O
scheme	O
,	O
we	O
define	O
a	O
model	O
of	O
ordinal	B-KEY
utility	I-KEY
,	O
which	O
in	O
turn	O
allows	O
us	O
to	O
interpret	O
our	O
Kripke-based	O
normative	B-KEY
systems	I-KEY
as	O
games	B-KEY
,	O
in	O
which	O
agents	O
must	O
determine	O
whether	O
to	O
comply	O
with	O
the	O
normative	B-KEY
system	I-KEY
or	O
not	O
.	O
We	O
then	O
characterise	O
the	O
computational	B-KEY
complexity	I-KEY
of	O
a	O
number	O
of	O
decision	O
problems	O
associated	O
with	O
these	O
Kripke-based	O
normative	O
system	O
games	O
;	O
for	O
example	O
,	O
we	O
show	O
that	O
the	O
complexity	O
of	O
checking	O
whether	O
there	O
exists	O
a	O
normative	O
system	O
which	O
has	O
the	O
property	O
of	O
being	O
a	O
Nash	O
implementation	O
is	O
NP-complete	O
.	O

Complexity	O
of	O
-LRB-	O
Iterated	O
-RRB-	O
Dominance	B-KEY
*	O
ABSTRACT	O
We	O
study	O
various	O
computational	O
aspects	O
of	O
solving	O
games	O
using	O
dominance	B-KEY
and	O
iterated	B-KEY
dominance	I-KEY
.	O
We	O
first	O
study	O
both	O
strict	O
and	O
weak	O
dominance	B-KEY
-LRB-	O
not	O
iterated	O
-RRB-	O
,	O
and	O
show	O
that	O
checking	O
whether	O
a	O
given	O
strategy	B-KEY
is	O
dominated	B-KEY
by	O
some	O
mixed	O
strategy	B-KEY
can	O
be	O
done	O
in	O
polynomial	O
time	O
using	O
a	O
single	O
linear	O
program	O
solve	O
.	O
We	O
then	O
move	O
on	O
to	O
iterated	B-KEY
dominance	I-KEY
.	O
We	O
show	O
that	O
determining	O
whether	O
there	O
is	O
some	O
path	O
that	O
eliminates	B-KEY
a	O
given	O
strategy	B-KEY
is	O
NP-complete	O
with	O
iterated	O
weak	O
dominance	B-KEY
.	O
This	O
allows	O
us	O
to	O
also	O
show	O
that	O
determining	O
whether	O
there	O
is	O
a	O
path	O
that	O
leads	O
to	O
a	O
unique	O
solution	O
is	O
NP-complete	O
.	O
Both	O
of	O
these	O
results	O
hold	O
both	O
with	O
and	O
without	O
dominance	B-KEY
by	O
mixed	O
strategies	B-KEY
.	O
-LRB-	O
A	O
weaker	O
version	O
of	O
the	O
second	O
result	O
-LRB-	O
only	O
without	O
dominance	B-KEY
by	O
mixed	O
strategies	B-KEY
-RRB-	O
was	O
already	O
known	O
-LSB-	O
7	O
-RSB-	O
.	O
-RRB-	O
Iterated	O
strict	O
dominance	B-KEY
,	O
on	O
the	O
other	O
hand	O
,	O
is	O
path-independent	O
-LRB-	O
both	O
with	O
and	O
without	O
dominance	B-KEY
by	O
mixed	O
strategies	B-KEY
-RRB-	O
and	O
can	O
therefore	O
be	O
done	O
in	O
polynomial	O
time	O
.	O
We	O
then	O
study	O
what	O
happens	O
when	O
the	O
dominating	B-KEY
strategy	B-KEY
is	O
allowed	O
to	O
place	O
positive	O
probability	O
on	O
only	O
a	O
few	O
pure	O
strategies	B-KEY
.	O
First	O
,	O
we	O
show	O
that	O
finding	O
the	O
dominating	B-KEY
strategy	B-KEY
with	O
minimum	O
support	O
size	O
is	O
NP-complete	O
-LRB-	O
both	O
for	O
strict	O
and	O
weak	O
dominance	B-KEY
-RRB-	O
.	O
Then	O
,	O
we	O
show	O
that	O
iterated	O
strict	O
dominance	B-KEY
becomes	O
path-dependent	O
when	O
there	O
is	O
a	O
limit	O
on	O
the	O
support	O
size	O
of	O
the	O
dominating	B-KEY
strategies	B-KEY
,	O
and	O
that	O
deciding	O
whether	O
a	O
given	O
strategy	B-KEY
can	O
be	O
eliminated	B-KEY
by	O
iterated	O
strict	O
dominance	B-KEY
under	O
this	O
restriction	O
is	O
NP-complete	O
-LRB-	O
even	O
when	O
the	O
limit	O
on	O
the	O
support	O
size	O
is	O
3	O
-RRB-	O
.	O
Finally	O
,	O
we	O
study	O
Bayesian	B-KEY
games	I-KEY
.	O
We	O
show	O
that	O
,	O
unlike	O
in	O
normal	B-KEY
form	I-KEY
games	I-KEY
,	O
deciding	O
whether	O
a	O
given	O
pure	O
strategy	B-KEY
is	O
dominated	B-KEY
by	O
another	O
pure	O
strategy	B-KEY
in	O
a	O
Bayesian	B-KEY
game	I-KEY
is	O
NP-complete	O
-LRB-	O
both	O
with	O
strict	O
and	O
weak	O
dominance	B-KEY
-RRB-	O
;	O
however	O
,	O
deciding	O
whether	O
a	O
strategy	B-KEY
is	O
dominated	B-KEY
by	O
some	O
mixed	O
strategy	B-KEY
can	O
still	O
be	O
done	O
in	O
polynomial	O
time	O
with	O
a	O
single	O
linear	O
program	O
solve	O
-LRB-	O
both	O
with	O
strict	O
and	O
weak	O
*	O
This	O
material	O
is	O
based	O
upon	O
work	O
supported	O
by	O
the	O
National	O
Science	O
Foundation	O
under	O
ITR	O
grants	O
IIS-0121678	O
and	O
IIS-0427858	O
,	O
and	O
a	O
Sloan	O
Fellowship	O
.	O
dominance	B-KEY
-RRB-	O
.	O
Finally	O
,	O
we	O
show	O
that	O
iterated	B-KEY
dominance	I-KEY
using	O
pure	O
strategies	O
can	O
require	O
an	O
exponential	O
number	O
of	O
iterations	O
in	O
a	O
Bayesian	O
game	O
-LRB-	O
both	O
with	O
strict	O
and	O
weak	O
dominance	O
-RRB-	O
.	O

A	O
Scalable	O
Distributed	O
Information	B-KEY
Management	I-KEY
System	I-KEY
*	O
We	O
present	O
a	O
Scalable	O
Distributed	O
Information	B-KEY
Management	I-KEY
System	I-KEY
-LRB-	O
SDIMS	O
-RRB-	O
that	O
aggregates	O
information	O
about	O
large-scale	O
networked	O
systems	O
and	O
that	O
can	O
serve	O
as	O
a	O
basic	O
building	O
block	O
for	O
a	O
broad	O
range	O
of	O
large-scale	O
distributed	O
applications	O
by	O
providing	O
detailed	O
views	O
of	O
nearby	O
information	O
and	O
summary	O
views	O
of	O
global	O
information	O
.	O
To	O
serve	O
as	O
a	O
basic	O
building	O
block	O
,	O
a	O
SDIMS	O
should	O
have	O
four	O
properties	O
:	O
scalability	O
to	O
many	O
nodes	O
and	O
attributes	O
,	O
flexibility	O
to	O
accommodate	O
a	O
broad	O
range	O
of	O
applications	O
,	O
administrative	B-KEY
isolation	I-KEY
for	O
security	O
and	O
availability	B-KEY
,	O
and	O
robustness	O
to	O
node	O
and	O
network	O
failures	O
.	O
We	O
design	O
,	O
implement	O
and	O
evaluate	O
a	O
SDIMS	O
that	O
-LRB-	O
1	O
-RRB-	O
leverages	O
Distributed	B-KEY
Hash	I-KEY
Tables	I-KEY
-LRB-	O
DHT	O
-RRB-	O
to	O
create	O
scalable	O
aggregation	O
trees	O
,	O
-LRB-	O
2	O
-RRB-	O
provides	O
flexibility	O
through	O
a	O
simple	O
API	O
that	O
lets	O
applications	O
control	O
propagation	O
of	O
reads	O
and	O
writes	O
,	O
-LRB-	O
3	O
-RRB-	O
provides	O
administrative	B-KEY
isolation	I-KEY
through	O
simple	O
extensions	O
to	O
current	O
DHT	O
algorithms	O
,	O
and	O
-LRB-	O
4	O
-RRB-	O
achieves	O
robustness	O
to	O
node	O
and	O
network	O
reconfigurations	O
through	O
lazy	O
reaggregation	O
,	O
on-demand	O
reaggregation	O
,	O
and	O
tunable	B-KEY
spatial	I-KEY
replication	I-KEY
.	O
Through	O
extensive	O
simulations	O
and	O
micro-benchmark	O
experiments	O
,	O
we	O
observe	O
that	O
our	O
system	O
is	O
an	O
order	O
of	O
magnitude	O
more	O
scalable	O
than	O
existing	O
approaches	O
,	O
achieves	O
isolation	O
properties	O
at	O
the	O
cost	O
of	O
modestly	O
increased	O
read	O
latency	O
in	O
comparison	O
to	O
flat	O
DHTs	O
,	O
and	O
gracefully	O
handles	O
failures	O
.	O

Efficiency	O
and	O
Nash	O
Equilibria	O
in	O
a	O
Scrip	B-KEY
System	I-KEY
for	O
P2P	B-KEY
Networks	I-KEY
ABSTRACT	O
A	O
model	O
of	O
providing	O
service	O
in	O
a	O
P2P	B-KEY
network	I-KEY
is	O
analyzed	O
.	O
It	O
is	O
shown	O
that	O
by	O
adding	O
a	O
scrip	B-KEY
system	I-KEY
,	O
a	O
mechanism	O
that	O
admits	O
a	O
reasonable	O
Nash	B-KEY
equilibrium	I-KEY
that	O
reduces	O
free	O
riding	O
can	O
be	O
obtained	O
.	O
The	O
effect	O
of	O
varying	O
the	O
total	O
amount	O
of	O
money	O
-LRB-	O
scrip	O
-RRB-	O
in	O
the	O
system	O
on	O
efficiency	O
-LRB-	O
i.e.	O
,	O
social	B-KEY
welfare	I-KEY
-RRB-	O
is	O
analyzed	O
,	O
and	O
it	O
is	O
shown	O
that	O
by	O
maintaining	O
the	O
appropriate	O
ratio	O
between	O
the	O
total	O
amount	O
of	O
money	O
and	O
the	O
number	O
of	O
agents	B-KEY
,	O
efficiency	O
is	O
maximized	O
.	O
The	O
work	O
has	O
implications	O
for	O
many	O
online	B-KEY
systems	I-KEY
,	O
not	O
only	O
P2P	B-KEY
networks	I-KEY
but	O
also	O
a	O
wide	O
variety	O
of	O
online	O
forums	O
for	O
which	O
scrip	B-KEY
systems	I-KEY
are	O
popular	O
,	O
but	O
formal	O
analyses	O
have	O
been	O
lacking	O
.	O

Rumours	O
and	O
Reputation	O
:	O
Evaluating	O
Multi-Dimensional	B-KEY
Trust	I-KEY
within	O
a	O
Decentralised	O
Reputation	B-KEY
System	I-KEY
ABSTRACT	O
In	O
this	O
paper	O
we	O
develop	O
a	O
novel	O
probabilistic	O
model	O
of	O
computational	O
trust	O
that	O
explicitly	O
deals	O
with	O
correlated	B-KEY
multi-dimensional	O
contracts	O
.	O
Our	O
starting	O
point	O
is	O
to	O
consider	O
an	O
agent	O
attempting	O
to	O
estimate	O
the	O
utility	O
of	O
a	O
contract	O
,	O
and	O
we	O
show	O
that	O
this	O
leads	O
to	O
a	O
model	O
of	O
computational	O
trust	O
whereby	O
an	O
agent	O
must	O
determine	O
a	O
vector	O
of	O
estimates	O
that	O
represent	O
the	O
probability	O
that	O
any	O
dimension	O
of	O
the	O
contract	O
will	O
be	O
successfully	O
fulfilled	O
,	O
and	O
a	O
covariance	O
matrix	O
that	O
describes	O
the	O
uncertainty	O
and	O
correlations	B-KEY
in	O
these	O
probabilities	O
.	O
We	O
present	O
a	O
formalism	O
based	O
on	O
the	O
Dirichlet	B-KEY
distribution	I-KEY
that	O
allows	O
an	O
agent	O
to	O
calculate	O
these	O
probabilities	O
and	O
correlations	B-KEY
from	O
their	O
direct	O
experience	O
of	O
contract	O
outcomes	O
,	O
and	O
we	O
show	O
that	O
this	O
leads	O
to	O
superior	O
estimates	O
compared	O
to	O
an	O
alternative	O
approach	O
using	O
multiple	O
independent	O
beta	O
distributions	O
.	O
We	O
then	O
show	O
how	O
agents	O
may	O
use	O
the	O
sufficient	O
statistics	O
of	O
this	O
Dirichlet	B-KEY
distribution	I-KEY
to	O
communicate	O
and	O
fuse	O
reputation	O
within	O
a	O
decentralised	O
reputation	B-KEY
system	I-KEY
.	O
Finally	O
,	O
we	O
present	O
a	O
novel	O
solution	O
to	O
the	O
problem	O
of	O
rumour	B-KEY
propagation	I-KEY
within	O
such	O
systems	O
.	O
This	O
solution	O
uses	O
the	O
notion	O
of	O
private	O
and	O
shared	O
information	O
,	O
and	O
provides	O
estimates	O
consistent	O
with	O
a	O
centralised	O
reputation	B-KEY
system	I-KEY
,	O
whilst	O
maintaining	O
the	O
anonymity	B-KEY
of	O
the	O
agents	O
,	O
and	O
avoiding	O
bias	O
and	O
overconfidence	B-KEY
.	O

Relaxed	O
Online	O
SVMs	O
for	O
Spam	B-KEY
Filtering	I-KEY
ABSTRACT	O
Spam	O
is	O
a	O
key	O
problem	O
in	O
electronic	O
communication	O
,	O
including	O
large-scale	O
email	O
systems	O
and	O
the	O
growing	O
number	O
of	O
blogs	B-KEY
.	O
Content-based	O
filtering	O
is	O
one	O
reliable	O
method	O
of	O
combating	O
this	O
threat	O
in	O
its	O
various	O
forms	O
,	O
but	O
some	O
academic	O
researchers	O
and	O
industrial	O
practitioners	O
disagree	O
on	O
how	O
best	O
to	O
filter	O
spam	O
.	O
The	O
former	O
have	O
advocated	O
the	O
use	O
of	O
Support	B-KEY
Vector	I-KEY
Machines	I-KEY
-LRB-	O
SVMs	O
-RRB-	O
for	O
content-based	O
filtering	O
,	O
as	O
this	O
machine	O
learning	O
methodology	O
gives	O
state-of-the-art	O
performance	O
for	O
text	O
classification	O
.	O
However	O
,	O
similar	O
performance	O
gains	O
have	O
yet	O
to	O
be	O
demonstrated	O
for	O
online	O
spam	B-KEY
filtering	I-KEY
.	O
Additionally	O
,	O
practitioners	O
cite	O
the	O
high	O
cost	O
of	O
SVMs	O
as	O
reason	O
to	O
prefer	O
faster	O
-LRB-	O
if	O
less	O
statistically	O
robust	O
-RRB-	O
Bayesian	B-KEY
methods	I-KEY
.	O
In	O
this	O
paper	O
,	O
we	O
offer	O
a	O
resolution	O
to	O
this	O
controversy	O
.	O
First	O
,	O
we	O
show	O
that	O
online	O
SVMs	O
indeed	O
give	O
state-of-the-art	O
classification	O
performance	O
on	O
online	O
spam	B-KEY
filtering	I-KEY
on	O
large	O
benchmark	O
data	O
sets	O
.	O
Second	O
,	O
we	O
show	O
that	O
nearly	O
equivalent	O
performance	O
may	O
be	O
achieved	O
by	O
a	O
Relaxed	O
Online	O
SVM	O
-LRB-	O
ROSVM	O
-RRB-	O
at	O
greatly	O
reduced	O
computational	O
cost	O
.	O
Our	O
results	O
are	O
experimentally	O
verified	O
on	O
email	O
spam	O
,	O
blog	B-KEY
spam	O
,	O
and	O
splog	B-KEY
detection	O
tasks	O
.	O

Dynamics	B-KEY
Based	I-KEY
Control	I-KEY
with	O
an	O
Application	O
to	O
Area-Sweeping	O
Problems	O
ABSTRACT	O
In	O
this	O
paper	O
we	O
introduce	O
Dynamics	B-KEY
Based	I-KEY
Control	I-KEY
-LRB-	O
DBC	O
-RRB-	O
,	O
an	O
approach	O
to	O
planning	O
and	O
control	O
of	O
an	O
agent	O
in	O
stochastic	O
environments	O
.	O
Unlike	O
existing	O
approaches	O
,	O
which	O
seek	O
to	O
optimize	O
expected	O
rewards	O
-LRB-	O
e.g.	O
,	O
in	O
Partially	B-KEY
Observable	I-KEY
Markov	I-KEY
Decision	I-KEY
Problems	I-KEY
-LRB-	O
POMDPs	O
-RRB-	O
-RRB-	O
,	O
DBC	O
optimizes	O
system	O
behavior	O
towards	O
specified	O
system	B-KEY
dynamics	I-KEY
.	O
We	O
show	O
that	O
a	O
recently	O
developed	O
planning	O
and	O
control	B-KEY
approach	O
,	O
Extended	B-KEY
Markov	I-KEY
Tracking	I-KEY
-LRB-	O
EMT	O
-RRB-	O
is	O
an	O
instantiation	O
of	O
DBC	O
.	O
EMT	O
employs	O
greedy	O
action	O
selection	O
to	O
provide	O
an	O
efficient	O
control	B-KEY
algorithm	O
in	O
Markovian	O
environments	O
.	O
We	O
exploit	O
this	O
efficiency	O
in	O
a	O
set	O
of	O
experiments	O
that	O
applied	O
multitarget	O
EMT	O
to	O
a	O
class	O
of	O
area-sweeping	B-KEY
problems	I-KEY
-LRB-	O
searching	O
for	O
moving	O
targets	O
-RRB-	O
.	O
We	O
show	O
that	O
such	O
problems	O
can	O
be	O
naturally	O
defined	O
and	O
efficiently	O
solved	O
using	O
the	O
DBC	O
framework	O
,	O
and	O
its	O
EMT	O
instantiation	O
.	O

Improving	O
Web	B-KEY
Search	I-KEY
Ranking	O
by	O
Incorporating	O
User	O
Behavior	O
Information	O
ABSTRACT	O
We	O
show	O
that	O
incorporating	O
user	B-KEY
behavior	I-KEY
data	O
can	O
significantly	O
improve	O
ordering	O
of	O
top	O
results	B-KEY
in	O
real	O
web	B-KEY
search	I-KEY
setting	O
.	O
We	O
examine	O
alternatives	O
for	O
incorporating	O
feedback	B-KEY
into	O
the	O
ranking	B-KEY
process	O
and	O
explore	O
the	O
contributions	O
of	O
user	O
feedback	B-KEY
compared	O
to	O
other	O
common	O
web	B-KEY
search	I-KEY
features	O
.	O
We	O
report	O
results	B-KEY
of	O
a	O
large	O
scale	O
evaluation	O
over	O
3,000	O
queries	O
and	O
12	O
million	O
user	B-KEY
interactions	I-KEY
with	O
a	O
popular	O
web	B-KEY
search	I-KEY
engine	O
.	O
We	O
show	O
that	O
incorporating	O
implicit	O
feedback	B-KEY
can	O
augment	O
other	O
features	O
,	O
improving	O
the	O
accuracy	O
of	O
a	O
competitive	O
web	B-KEY
search	I-KEY
ranking	O
algorithms	O
by	O
as	O
much	O
as	O
31	O
%	O
relative	O
to	O
the	O
original	O
performance	O
.	O

Downloading	O
Textual	O
Hidden	B-KEY
Web	I-KEY
Content	O
Through	O
Keyword	B-KEY
Queries	I-KEY
ABSTRACT	O
An	O
ever-increasing	O
amount	O
of	O
information	O
on	O
the	O
Web	O
today	O
is	O
available	O
only	O
through	O
search	O
interfaces	O
:	O
the	O
users	O
have	O
to	O
type	O
in	O
a	O
set	O
of	O
keywords	O
in	O
a	O
search	O
form	O
in	O
order	O
to	O
access	O
the	O
pages	O
from	O
certain	O
Web	O
sites	O
.	O
These	O
pages	O
are	O
often	O
referred	O
to	O
as	O
the	O
Hidden	B-KEY
Web	I-KEY
or	O
the	O
Deep	B-KEY
Web	I-KEY
.	O
Since	O
there	O
are	O
no	O
static	O
links	O
to	O
the	O
Hidden	B-KEY
Web	I-KEY
pages	O
,	O
search	O
engines	O
can	O
not	O
discover	O
and	O
index	O
such	O
pages	O
and	O
thus	O
do	O
not	O
return	O
them	O
in	O
the	O
results	O
.	O
However	O
,	O
according	O
to	O
recent	O
studies	O
,	O
the	O
content	O
provided	O
by	O
many	O
Hidden	B-KEY
Web	I-KEY
sites	O
is	O
often	O
of	O
very	O
high	O
quality	O
and	O
can	O
be	O
extremely	O
valuable	O
to	O
many	O
users	O
.	O
In	O
this	O
paper	O
,	O
we	O
study	O
how	O
we	O
can	O
build	O
an	O
effective	O
Hidden	B-KEY
Web	I-KEY
crawler	O
that	O
can	O
autonomously	O
discover	O
and	O
download	O
pages	O
from	O
the	O
Hidden	O
Web	O
.	O
Since	O
the	O
only	O
``	O
entry	O
point	O
''	O
to	O
a	O
Hidden	B-KEY
Web	I-KEY
site	O
is	O
a	O
query	O
interface	O
,	O
the	O
main	O
challenge	O
that	O
a	O
Hidden	B-KEY
Web	I-KEY
crawler	O
has	O
to	O
face	O
is	O
how	O
to	O
automatically	O
generate	O
meaningful	O
queries	O
to	O
issue	O
to	O
the	O
site	O
.	O
Here	O
,	O
we	O
provide	O
a	O
theoretical	O
framework	O
to	O
investigate	O
the	O
query	O
generation	O
problem	O
for	O
the	O
Hidden	B-KEY
Web	I-KEY
and	O
we	O
propose	O
effective	O
policies	O
for	O
generating	O
queries	O
automatically	O
.	O
Our	O
policies	O
proceed	O
iteratively	O
,	O
issuing	O
a	O
different	O
query	O
in	O
every	O
iteration	O
.	O
We	O
experimentally	O
evaluate	O
the	O
effectiveness	O
of	O
these	O
policies	O
on	O
4	O
real	O
Hidden	B-KEY
Web	I-KEY
sites	O
and	O
our	O
results	O
are	O
very	O
promising	O
.	O
For	O
instance	O
,	O
in	O
one	O
experiment	O
,	O
one	O
of	O
our	O
policies	O
downloaded	O
more	O
than	O
90	O
%	O
of	O
a	O
Hidden	B-KEY
Web	I-KEY
site	O
-LRB-	O
that	O
contains	O
14	O
million	O
documents	O
-RRB-	O
after	O
issuing	O
fewer	O
than	O
100	O
queries	O
.	O

Unified	O
Utility	O
Maximization	O
Framework	O
for	O
Resource	B-KEY
Selection	I-KEY
ABSTRACT	O
This	O
paper	O
presents	O
a	O
unified	O
utility	O
framework	O
for	O
resource	B-KEY
selection	I-KEY
of	O
distributed	O
text	O
information	O
retrieval	O
.	O
This	O
new	O
framework	O
shows	O
an	O
efficient	O
and	O
effective	O
way	O
to	O
infer	O
the	O
probabilities	O
of	O
relevance	O
of	O
all	O
the	O
documents	O
across	O
the	O
text	O
databases	O
.	O
With	O
the	O
estimated	O
relevance	O
information	O
,	O
resource	B-KEY
selection	I-KEY
can	O
be	O
made	O
by	O
explicitly	O
optimizing	O
the	O
goals	O
of	O
different	O
applications	O
.	O
Specifically	O
,	O
when	O
used	O
for	O
database	B-KEY
recommendation	I-KEY
,	O
the	O
selection	O
is	O
optimized	O
for	O
the	O
goal	O
of	O
highrecall	O
-LRB-	O
include	O
as	O
many	O
relevant	O
documents	O
as	O
possible	O
in	O
the	O
selected	O
databases	O
-RRB-	O
;	O
when	O
used	O
for	O
distributed	B-KEY
document	I-KEY
retrieval	I-KEY
,	O
the	O
selection	O
targets	O
the	O
high-precision	O
goal	O
-LRB-	O
high	O
precision	O
in	O
the	O
final	O
merged	O
list	O
of	O
documents	O
-RRB-	O
.	O
This	O
new	O
model	O
provides	O
a	O
more	O
solid	O
framework	O
for	O
distributed	B-KEY
information	I-KEY
retrieval	I-KEY
.	O
Empirical	O
studies	O
show	O
that	O
it	O
is	O
at	O
least	O
as	O
effective	O
as	O
other	O
state-of-the-art	O
algorithms	O
.	O

Performance	B-KEY
Prediction	I-KEY
Using	O
Spatial	B-KEY
Autocorrelation	I-KEY
ABSTRACT	O
Evaluation	O
of	O
information	B-KEY
retrieval	I-KEY
systems	O
is	O
one	O
of	O
the	O
core	O
tasks	O
in	O
information	B-KEY
retrieval	I-KEY
.	O
Problems	O
include	O
the	O
inability	O
to	O
exhaustively	O
label	O
all	O
documents	O
for	O
a	O
topic	O
,	O
nongeneralizability	O
from	O
a	O
small	O
number	O
of	O
topics	O
,	O
and	O
incorporating	O
the	O
variability	O
of	O
retrieval	O
systems	O
.	O
Previous	O
work	O
addresses	O
the	O
evaluation	O
of	O
systems	O
,	O
the	O
ranking	B-KEY
of	I-KEY
queries	I-KEY
by	O
difficulty	O
,	O
and	O
the	O
ranking	O
of	O
individual	O
retrievals	O
by	O
performance	O
.	O
Approaches	O
exist	O
for	O
the	O
case	O
of	O
few	O
and	O
even	O
no	O
relevance	O
judgments	O
.	O
Our	O
focus	O
is	O
on	O
zero-judgment	O
performance	B-KEY
prediction	I-KEY
of	O
individual	O
retrievals	O
.	O
One	O
common	O
shortcoming	O
of	O
previous	O
techniques	O
is	O
the	O
assumption	O
of	O
uncorrelated	O
document	O
scores	O
and	O
judgments	O
.	O
If	O
documents	O
are	O
embedded	O
in	O
a	O
high-dimensional	O
space	O
-LRB-	O
as	O
they	O
often	O
are	O
-RRB-	O
,	O
we	O
can	O
apply	O
techniques	O
from	O
spatial	O
data	O
analysis	O
to	O
detect	O
correlations	O
between	O
document	O
scores	O
.	O
We	O
find	O
that	O
the	O
low	O
correlation	O
between	O
scores	O
of	O
topically	O
close	O
documents	O
often	O
implies	O
a	O
poor	O
retrieval	O
performance	O
.	O
When	O
compared	O
to	O
a	O
state	O
of	O
the	O
art	O
baseline	O
,	O
we	O
demonstrate	O
that	O
the	O
spatial	O
analysis	O
of	O
retrieval	O
scores	O
provides	O
significantly	O
better	O
prediction	O
performance	O
.	O
These	O
new	O
predictors	O
can	O
also	O
be	O
incorporated	O
with	O
classic	O
predictors	O
to	O
improve	O
performance	O
further	O
.	O
We	O
also	O
describe	O
the	O
first	O
large-scale	O
experiment	O
to	O
evaluate	O
zero-judgment	O
performance	B-KEY
prediction	I-KEY
for	O
a	O
massive	O
number	O
of	O
retrieval	O
systems	O
over	O
a	O
variety	O
of	O
collections	O
in	O
several	O
languages	O
.	O

Learning	O
User	O
Interaction	O
Models	O
for	O
Predicting	O
Web	O
Search	O
Result	O
Preferences	O
ABSTRACT	O
Evaluating	O
user	B-KEY
preferences	I-KEY
of	O
web	O
search	O
results	O
is	O
crucial	O
for	O
search	O
engine	O
development	O
,	O
deployment	O
,	O
and	O
maintenance	O
.	O
We	O
present	O
a	O
real-world	O
study	O
of	O
modeling	O
the	O
behavior	O
of	O
web	O
search	O
users	O
to	O
predict	O
web	O
search	O
result	O
preferences	O
.	O
Accurate	O
modeling	O
and	O
interpretation	O
of	O
user	O
behavior	O
has	O
important	O
applications	O
to	O
ranking	O
,	O
click	B-KEY
spam	I-KEY
detection	I-KEY
,	O
web	O
search	O
personalization	B-KEY
,	O
and	O
other	O
tasks	O
.	O
Our	O
key	O
insight	O
to	O
improving	O
robustness	O
of	O
interpreting	O
implicit	B-KEY
feedback	I-KEY
is	O
to	O
model	O
query-dependent	O
deviations	O
from	O
the	O
expected	O
``	O
noisy	O
''	O
user	O
behavior	O
.	O
We	O
show	O
that	O
our	O
model	O
of	O
clickthrough	B-KEY
interpretation	O
improves	O
prediction	O
accuracy	O
over	O
state-of-the-art	O
clickthrough	B-KEY
methods	O
.	O
We	O
generalize	O
our	O
approach	O
to	O
model	O
user	O
behavior	O
beyond	O
clickthrough	B-KEY
,	O
which	O
results	O
in	O
higher	O
preference	O
prediction	O
accuracy	O
than	O
models	O
based	O
on	O
clickthrough	B-KEY
information	O
alone	O
.	O
We	O
report	O
results	O
of	O
a	O
large-scale	O
experimental	O
evaluation	O
that	O
show	O
substantial	O
improvements	O
over	O
published	O
implicit	B-KEY
feedback	I-KEY
interpretation	O
methods	O
.	O

Robust	B-KEY
Solutions	O
for	O
Combinatorial	B-KEY
Auctions	I-KEY
*	O
ABSTRACT	O
Bids	B-KEY
submitted	O
in	O
auctions	O
are	O
usually	O
treated	O
as	O
enforceable	B-KEY
commitments	I-KEY
in	O
most	O
bidding	B-KEY
and	O
auction	O
theory	O
literature	O
.	O
In	O
reality	O
bidders	O
often	O
withdraw	O
winning	O
bids	B-KEY
before	O
the	O
transaction	O
when	O
it	O
is	O
in	O
their	O
best	O
interests	O
to	O
do	O
so	O
.	O
Given	O
a	O
bid	B-KEY
withdrawal	O
in	O
a	O
combinatorial	O
auction	O
,	O
finding	O
an	O
alternative	O
repair	O
solution	O
of	O
adequate	O
revenue	O
without	O
causing	O
undue	O
disturbance	O
to	O
the	O
remaining	O
winning	O
bids	O
in	O
the	O
original	O
solution	O
may	O
be	O
difficult	O
or	O
even	O
impossible	O
.	O
We	O
have	O
called	O
this	O
the	O
``	O
Bid-taker	O
's	O
Exposure	O
Problem	O
''	O
.	O
When	O
faced	O
with	O
such	O
unreliable	O
bidders	O
,	O
it	O
is	O
preferable	O
for	O
the	O
bid-taker	O
to	O
preempt	O
such	O
uncertainty	O
by	O
having	O
a	O
solution	O
that	O
is	O
robust	O
to	O
bid	O
withdrawal	O
and	O
provides	O
a	O
guarantee	O
that	O
possible	O
withdrawals	O
may	O
be	O
repaired	O
easily	O
with	O
a	O
bounded	O
loss	O
in	O
revenue	O
.	O
In	O
this	O
paper	O
,	O
we	O
propose	O
an	O
approach	O
to	O
addressing	O
the	O
Bidtaker	O
's	O
Exposure	B-KEY
Problem	I-KEY
.	O
Firstly	O
,	O
we	O
use	O
the	O
Weighted	B-KEY
Super	I-KEY
Solutions	I-KEY
framework	O
-LSB-	O
13	O
-RSB-	O
,	O
from	O
the	O
field	O
of	O
constraint	B-KEY
programming	I-KEY
,	O
to	O
solve	O
the	O
problem	O
of	O
finding	O
a	O
robust	B-KEY
solution	O
.	O
A	O
weighted	B-KEY
super	I-KEY
solution	I-KEY
guarantees	O
that	O
any	O
subset	O
of	O
bids	B-KEY
likely	O
to	O
be	O
withdrawn	O
can	O
be	O
repaired	O
to	O
form	O
a	O
new	O
solution	O
of	O
at	O
least	O
a	O
given	O
revenue	O
by	O
making	O
limited	O
changes	O
.	O
Secondly	O
,	O
we	O
introduce	O
an	O
auction	O
model	O
that	O
uses	O
a	O
form	O
of	O
leveled	O
commitment	O
contract	O
-LSB-	O
26	O
,	O
27	O
-RSB-	O
,	O
which	O
we	O
have	O
called	O
mutual	O
bid	B-KEY
bonds	O
,	O
to	O
improve	O
solution	O
reparability	O
by	O
facilitating	O
backtracking	O
on	O
winning	O
bids	O
by	O
the	O
bid-taker	O
.	O
We	O
then	O
examine	O
the	O
trade-off	O
between	O
robustness	B-KEY
and	O
revenue	O
in	O
different	O
economically	O
motivated	O
auction	O
scenarios	O
for	O
different	O
constraints	O
on	O
the	O
revenue	O
of	O
repair	O
solutions	O
.	O
We	O
also	O
demonstrate	O
experimentally	O
that	O
fewer	O
winning	O
bids	B-KEY
partake	O
in	O
robust	B-KEY
solutions	O
,	O
thereby	O
reducing	O
any	O
associated	O
overhead	O
in	O
dealing	O
with	O
extra	O
bidders	O
.	O
Robust	B-KEY
solutions	O
can	O
also	O
provide	O
a	O
means	O
of	O
selectively	O
discriminating	O
against	O
distrusted	O
bidders	O
in	O
a	O
measured	O
manner	O
.	O

HITS	B-KEY
Hits	B-KEY
TREC	B-KEY
--	O
Exploring	O
IR	B-KEY
Evaluation	I-KEY
Results	O
with	O
Network	B-KEY
Analysis	I-KEY
ABSTRACT	O
We	O
propose	O
a	O
novel	O
method	O
of	O
analysing	O
data	O
gathered	O
from	O
TREC	B-KEY
or	O
similar	O
information	B-KEY
retrieval	I-KEY
evaluation	I-KEY
experiments	I-KEY
.	O
We	O
define	O
two	O
normalized	O
versions	O
of	O
average	O
precision	O
,	O
that	O
we	O
use	O
to	O
construct	O
a	O
weighted	B-KEY
bipartite	I-KEY
graph	I-KEY
of	O
TREC	B-KEY
systems	O
and	O
topics	O
.	O
We	O
analyze	O
the	O
meaning	O
of	O
well	O
known	O
--	O
and	O
somewhat	O
generalized	O
--	O
indicators	O
from	O
social	B-KEY
network	I-KEY
analysis	I-KEY
on	O
the	O
Systems-Topics	O
graph	O
.	O
We	O
apply	O
this	O
method	O
to	O
an	O
analysis	O
of	O
TREC	B-KEY
8	O
data	O
;	O
among	O
the	O
results	O
,	O
we	O
find	O
that	O
authority	O
measures	O
systems	O
performance	O
,	O
that	O
hubness	O
of	O
topics	O
reveals	O
that	O
some	O
topics	O
are	O
better	O
than	O
others	O
at	O
distinguishing	O
more	O
or	O
less	O
effective	O
systems	O
,	O
that	O
with	O
current	O
measures	O
a	O
system	O
that	O
wants	O
to	O
be	O
effective	O
in	O
TREC	B-KEY
needs	O
to	O
be	O
effective	O
on	O
easy	O
topics	O
,	O
and	O
that	O
by	O
using	O
different	O
effectiveness	O
measures	O
this	O
is	O
no	O
longer	O
the	O
case	O
.	O

Learning	O
and	O
Joint	B-KEY
Deliberation	I-KEY
through	O
Argumentation	B-KEY
in	O
Multi-Agent	O
Systems	O
ABSTRACT	O
In	O
this	O
paper	O
we	O
will	O
present	O
an	O
argumentation	B-KEY
framework	O
for	O
learning	O
agents	O
-LRB-	O
AMAL	O
-RRB-	O
designed	O
for	O
two	O
purposes	O
:	O
-LRB-	O
1	O
-RRB-	O
for	O
joint	O
deliberation	O
,	O
and	O
-LRB-	O
2	O
-RRB-	O
for	O
learning	O
from	O
communication	O
.	O
The	O
AMAL	O
framework	O
is	O
completely	O
based	O
on	O
learning	O
from	O
examples	O
:	O
the	O
argument	B-KEY
preference	O
relation	O
,	O
the	O
argument	B-KEY
generation	O
policy	O
,	O
and	O
the	O
counterargument	O
generation	O
policy	O
are	O
case-based	O
techniques	O
.	O
For	O
join	O
deliberation	O
,	O
learning	B-KEY
agents	I-KEY
share	O
their	O
experience	O
by	O
forming	O
a	O
committee	O
to	O
decide	O
upon	O
some	O
joint	O
decision	O
.	O
We	O
experimentally	O
show	O
that	O
the	O
argumentation	B-KEY
among	O
committees	O
of	O
agents	O
improves	O
both	O
the	O
individual	O
and	O
joint	O
performance	O
.	O
For	O
learning	B-KEY
from	I-KEY
communication	I-KEY
,	O
an	O
agent	O
engages	O
into	O
arguing	O
with	O
other	O
agents	O
in	O
order	O
to	O
contrast	O
its	O
individual	O
hypotheses	O
and	O
receive	O
counterexamples	O
;	O
the	O
argumentation	B-KEY
process	O
improves	O
their	O
learning	O
scope	O
and	O
individual	O
performance	O
.	O

Expressive	B-KEY
Negotiation	I-KEY
over	O
Donations	B-KEY
to	I-KEY
Charities	I-KEY
∗	O
ABSTRACT	O
When	O
donating	O
money	O
to	O
a	O
-LRB-	O
say	O
,	O
charitable	O
-RRB-	O
cause	O
,	O
it	O
is	O
possible	O
to	O
use	O
the	O
contemplated	O
donation	O
as	O
negotiating	B-KEY
material	I-KEY
to	O
induce	O
other	O
parties	O
interested	O
in	O
the	O
charity	O
to	O
donate	O
more	O
.	O
Such	O
negotiation	O
is	O
usually	O
done	O
in	O
terms	O
of	O
matching	O
offers	O
,	O
where	O
one	O
party	O
promises	O
to	O
pay	O
a	O
certain	O
amount	O
if	O
others	O
pay	O
a	O
certain	O
amount	O
.	O
However	O
,	O
in	O
their	O
current	O
form	O
,	O
matching	O
offers	O
allow	O
for	O
only	O
limited	O
negotiation	O
.	O
For	O
one	O
,	O
it	O
is	O
not	O
immediately	O
clear	O
how	O
multiple	O
parties	O
can	O
make	O
matching	O
offers	O
at	O
the	O
same	O
time	O
without	O
creating	O
circular	O
dependencies	O
.	O
Also	O
,	O
it	O
is	O
not	O
immediately	O
clear	O
how	O
to	O
make	O
a	O
donation	O
conditional	O
on	O
other	O
donations	O
to	O
multiple	O
charities	O
,	O
when	O
the	O
donator	O
has	O
different	O
levels	O
of	O
appreciation	O
for	O
the	O
different	O
charities	O
.	O
In	O
both	O
these	O
cases	O
,	O
the	O
limited	O
expressiveness	O
of	O
matching	O
offers	O
causes	O
economic	O
loss	O
:	O
it	O
may	O
happen	O
that	O
an	O
arrangement	O
that	O
would	O
have	O
made	O
all	O
parties	O
-LRB-	O
donators	O
as	O
well	O
as	O
charities	O
-RRB-	O
better	O
off	O
can	O
not	O
be	O
expressed	O
in	O
terms	O
of	O
matching	O
offers	O
and	O
will	O
therefore	O
not	O
occur	O
.	O
In	O
this	O
paper	O
,	O
we	O
introduce	O
a	O
bidding	B-KEY
language	I-KEY
for	O
expressing	O
very	O
general	O
types	O
of	O
matching	O
offers	O
over	O
multiple	O
charities	O
.	O
We	O
formulate	O
the	O
corresponding	O
clearing	O
problem	O
-LRB-	O
deciding	O
how	O
much	O
each	O
bidder	O
pays	O
,	O
and	O
how	O
much	O
each	O
charity	O
receives	O
-RRB-	O
,	O
and	O
show	O
that	O
it	O
is	O
NP-complete	O
to	O
approximate	O
to	O
any	O
ratio	O
even	O
in	O
very	O
restricted	O
settings	O
.	O
We	O
give	O
a	O
mixed-integer	O
program	O
formulation	O
of	O
the	O
clearing	O
problem	O
,	O
and	O
show	O
that	O
for	O
concave	B-KEY
bids	I-KEY
,	O
the	O
program	O
reduces	O
to	O
a	O
linear	B-KEY
program	I-KEY
.	O
We	O
then	O
show	O
that	O
the	O
clearing	O
problem	O
for	O
a	O
subclass	O
of	O
concave	B-KEY
bids	I-KEY
is	O
at	O
least	O
as	O
hard	O
as	O
the	O
decision	O
variant	O
of	O
linear	B-KEY
programming	I-KEY
.	O
Subsequently	O
,	O
we	O
show	O
that	O
the	O
clearing	O
problem	O
is	O
much	O
easier	O
when	O
bids	O
are	O
quasilinear	B-KEY
--	O
for	O
surplus	O
,	O
the	O
problem	O
decomposes	O
across	O
charities	O
,	O
and	O
for	O
payment	O
maximization	O
,	O
a	O
greedy	O
approach	O
is	O
optimal	O
if	O
the	O
bids	O
are	O
concave	O
-LRB-	O
although	O
this	O
latter	O
problem	O
is	O
weakly	O
NP-complete	O
when	O
the	O
bids	O
are	O
not	O
concave	O
-RRB-	O
.	O
For	O
the	O
quasilinear	B-KEY
setting	O
,	O
we	O
study	O
the	O
mechanism	B-KEY
design	I-KEY
question	O
.	O
We	O
show	O
that	O
an	O
ex-post	O
efficient	O
mechanism	O
is	O
∗	O
Supported	O
by	O
NSF	O
under	O
CAREER	O
Award	O
IRI-9703122	O
,	O
Grant	O
IIS-9800994	O
,	O
ITR	O
IIS-0081246	O
,	O
and	O
ITR	O
IIS-0121678	O
.	O
impossible	O
even	O
with	O
only	O
one	O
charity	O
and	O
a	O
very	O
restricted	O
class	O
of	O
bids	O
.	O
We	O
also	O
show	O
that	O
there	O
may	O
be	O
benefits	O
to	O
linking	O
the	O
charities	O
from	O
a	O
mechanism	B-KEY
design	I-KEY
standpoint	O
.	O

Location	O
based	O
Indexing	B-KEY
Scheme	O
for	O
DAYS	O
ABSTRACT	O
Data	O
dissemination	O
through	O
wireless	B-KEY
channels	I-KEY
for	O
broadcasting	O
information	O
to	O
consumers	O
is	O
becoming	O
quite	O
common	O
.	O
Many	O
dissemination	O
schemes	O
have	O
been	O
proposed	O
but	O
most	O
of	O
them	O
push	O
data	O
to	O
wireless	B-KEY
channels	I-KEY
for	O
general	O
consumption	O
.	O
Push	O
based	O
broadcast	O
-LSB-	O
1	O
-RSB-	O
is	O
essentially	O
asymmetric	O
,	O
i.e.	O
,	O
the	O
volume	O
of	O
data	O
being	O
higher	O
from	O
the	O
server	O
to	O
the	O
users	O
than	O
from	O
the	O
users	O
back	O
to	O
the	O
server	O
.	O
Push	O
based	O
scheme	O
requires	O
some	O
indexing	B-KEY
which	O
indicates	O
when	O
the	O
data	O
will	O
be	O
broadcast	O
and	O
its	O
position	O
in	O
the	O
broadcast	O
.	O
Access	O
latency	O
and	O
tuning	O
time	O
are	O
the	O
two	O
main	O
parameters	O
which	O
may	O
be	O
used	O
to	O
evaluate	O
an	O
indexing	B-KEY
scheme	O
.	O
Two	O
of	O
the	O
important	O
indexing	B-KEY
schemes	O
proposed	O
earlier	O
were	O
tree	O
based	O
and	O
the	O
exponential	O
indexing	O
schemes	O
.	O
None	O
of	O
these	O
schemes	O
were	O
able	O
to	O
address	O
the	O
requirements	O
of	O
location	B-KEY
dependent	I-KEY
data	I-KEY
-LRB-	O
LDD	B-KEY
-RRB-	O
which	O
is	O
highly	O
desirable	O
feature	O
of	O
data	O
dissemination	O
.	O
In	O
this	O
paper	O
,	O
we	O
discuss	O
the	O
broadcast	O
of	O
LDD	B-KEY
in	O
our	O
project	O
DAta	O
in	O
Your	O
Space	O
-LRB-	O
DAYS	O
-RRB-	O
,	O
and	O
propose	O
a	O
scheme	O
for	O
indexing	B-KEY
LDD	B-KEY
.	O
We	O
argue	O
that	O
this	O
scheme	O
,	O
when	O
applied	O
to	O
LDD	B-KEY
,	O
significantly	O
improves	O
performance	O
in	O
terms	O
of	O
tuning	O
time	O
over	O
the	O
above	O
mentioned	O
schemes	O
.	O
We	O
prove	O
our	O
argument	O
with	O
the	O
help	O
of	O
simulation	O
results	O
.	O

Vocabulary	B-KEY
Independent	O
Spoken	B-KEY
Term	I-KEY
Detection	I-KEY
ABSTRACT	O
We	O
are	O
interested	O
in	O
retrieving	O
information	O
from	O
speech	O
data	O
like	O
broadcast	O
news	O
,	O
telephone	O
conversations	O
and	O
roundtable	O
meetings	O
.	O
Today	O
,	O
most	O
systems	O
use	O
large	O
vocabulary	B-KEY
continuous	O
speech	O
recognition	O
tools	O
to	O
produce	O
word	O
transcripts	O
;	O
the	O
transcripts	O
are	O
indexed	O
and	O
query	O
terms	O
are	O
retrieved	O
from	O
the	O
index	O
.	O
However	O
,	O
query	O
terms	O
that	O
are	O
not	O
part	O
of	O
the	O
recognizer	O
's	O
vocabulary	B-KEY
can	O
not	O
be	O
retrieved	O
,	O
and	O
the	O
recall	O
of	O
the	O
search	O
is	O
affected	O
.	O
In	O
addition	O
to	O
the	O
output	O
word	O
transcript	O
,	O
advanced	O
systems	O
provide	O
also	O
phonetic	B-KEY
transcripts	I-KEY
,	O
against	O
which	O
query	O
terms	O
can	O
be	O
matched	O
phonetically	O
.	O
Such	O
phonetic	B-KEY
transcripts	I-KEY
suffer	O
from	O
lower	O
accuracy	O
and	O
can	O
not	O
be	O
an	O
alternative	O
to	O
word	O
transcripts	O
.	O
We	O
present	O
a	O
vocabulary	B-KEY
independent	O
system	O
that	O
can	O
handle	O
arbitrary	O
queries	O
,	O
exploiting	O
the	O
information	O
provided	O
by	O
having	O
both	O
word	O
transcripts	O
and	O
phonetic	O
transcripts	O
.	O
A	O
speech	B-KEY
recognizer	I-KEY
generates	O
word	O
confusion	O
networks	O
and	O
phonetic	O
lattices	O
.	O
The	O
transcripts	O
are	O
indexed	O
for	O
query	O
processing	O
and	O
ranking	O
purpose	O
.	O
The	O
value	O
of	O
the	O
proposed	O
method	O
is	O
demonstrated	O
by	O
the	O
relative	O
high	O
performance	O
of	O
our	O
system	O
,	O
which	O
received	O
the	O
highest	O
overall	O
ranking	O
for	O
US	O
English	O
speech	O
data	O
in	O
the	O
recent	O
NIST	O
Spoken	B-KEY
Term	I-KEY
Detection	I-KEY
evaluation	O
-LSB-	O
1	O
-RSB-	O
.	O

Globally	O
Synchronized	O
Dead-Reckoning	B-KEY
with	O
Local	B-KEY
Lag	I-KEY
for	O
Continuous	O
Distributed	O
Multiplayer	B-KEY
Games	I-KEY
ABSTRACT	O
Dead-Reckoning	B-KEY
-LRB-	O
DR	O
-RRB-	O
is	O
an	O
effective	O
method	O
to	O
maintain	O
consistency	B-KEY
for	O
Continuous	O
Distributed	O
Multiplayer	B-KEY
Games	I-KEY
-LRB-	O
CDMG	O
-RRB-	O
.	O
Since	O
DR	O
can	O
filter	O
most	O
unnecessary	O
state	O
updates	O
and	O
improve	O
the	O
scalability	O
of	O
a	O
system	O
,	O
it	O
is	O
widely	O
used	O
in	O
commercial	O
CDMG	O
.	O
However	O
,	O
DR	O
can	O
not	O
maintain	O
high	O
consistency	B-KEY
,	O
and	O
this	O
constrains	O
its	O
application	O
in	O
highly	O
interactive	O
games	O
.	O
With	O
the	O
help	O
of	O
global	O
synchronization	O
,	O
DR	O
can	O
achieve	O
higher	O
consistency	B-KEY
,	O
but	O
it	O
still	O
can	O
not	O
eliminate	O
before	O
inconsistency	O
.	O
In	O
this	O
paper	O
,	O
a	O
method	O
named	O
Globally	O
Synchronized	O
DR	O
with	O
Local	B-KEY
Lag	I-KEY
-LRB-	O
GS-DR-LL	B-KEY
-RRB-	O
,	O
which	O
combines	O
local	B-KEY
lag	I-KEY
and	O
Globally	O
Synchronized	O
DR	O
-LRB-	O
GS-DR	O
-RRB-	O
,	O
is	O
presented	O
.	O
Performance	O
evaluation	O
shows	O
that	O
GS-DR-LL	B-KEY
can	O
effectively	O
decrease	O
before	O
inconsistency	O
,	O
and	O
the	O
effects	O
increase	O
with	O
the	O
lag	O
.	O

Feature	O
Representation	O
for	O
Effective	O
Action-Item	B-KEY
Detection	I-KEY
ABSTRACT	O
E-mail	B-KEY
users	O
face	O
an	O
ever-growing	O
challenge	O
in	O
managing	O
their	O
inboxes	O
due	O
to	O
the	O
growing	O
centrality	O
of	O
email	O
in	O
the	O
workplace	O
for	O
task	O
assignment	O
,	O
action	O
requests	O
,	O
and	O
other	O
roles	O
beyond	O
information	O
dissemination	O
.	O
Whereas	O
Information	B-KEY
Retrieval	I-KEY
and	O
Machine	O
Learning	O
techniques	O
are	O
gaining	O
initial	O
acceptance	O
in	O
spam	O
filtering	O
and	O
automated	O
folder	O
assignment	O
,	O
this	O
paper	O
reports	O
on	O
a	O
new	O
task	O
:	O
automated	O
action-item	B-KEY
detection	I-KEY
,	O
in	O
order	O
to	O
flag	O
emails	O
that	O
require	O
responses	O
,	O
and	O
to	O
highlight	O
the	O
specific	O
passage	O
-LRB-	O
s	O
-RRB-	O
indicating	O
the	O
request	O
-LRB-	O
s	O
-RRB-	O
for	O
action	O
.	O
Unlike	O
standard	O
topic-driven	O
text	B-KEY
classification	I-KEY
,	O
action-item	O
detection	O
requires	O
inferring	O
the	O
sender	O
's	O
intent	O
,	O
and	O
as	O
such	O
responds	O
less	O
well	O
to	O
pure	O
bag-of-words	O
classification	O
.	O
However	O
,	O
using	O
enriched	O
feature	O
sets	O
,	O
such	O
as	O
n-grams	B-KEY
-LRB-	O
up	O
to	O
n	O
=	O
4	O
-RRB-	O
with	O
chi-squared	O
feature	B-KEY
selection	I-KEY
,	O
and	O
contextual	O
cues	O
for	O
action-item	O
location	O
improve	O
performance	O
by	O
up	O
to	O
10	O
%	O
over	O
unigrams	O
,	O
using	O
in	O
both	O
cases	O
state	O
of	O
the	O
art	O
classifiers	O
such	O
as	O
SVMs	O
with	O
automated	O
model	O
selection	O
via	O
embedded	O
cross-validation	O
.	O

Distance	B-KEY
Measures	I-KEY
for	O
MPEG-7-based	O
Retrieval	O
ABSTRACT	O
In	O
visual	B-KEY
information	I-KEY
retrieval	I-KEY
the	O
careful	O
choice	O
of	O
suitable	O
proximity	O
measures	O
is	O
a	O
crucial	O
success	O
factor	O
.	O
The	O
evaluation	O
presented	O
in	O
this	O
paper	O
aims	O
at	O
showing	O
that	O
the	O
distance	B-KEY
measures	I-KEY
suggested	O
by	O
the	O
MPEG-7	B-KEY
group	O
for	O
the	O
visual	B-KEY
descriptors	I-KEY
can	O
be	O
beaten	O
by	O
general-purpose	O
measures	O
.	O
Eight	O
visual	O
MPEG-7	B-KEY
descriptors	O
were	O
selected	O
and	O
38	O
distance	B-KEY
measures	I-KEY
implemented	O
.	O
Three	O
media	B-KEY
collections	I-KEY
were	O
created	O
and	O
assessed	O
,	O
performance	B-KEY
indicators	I-KEY
developed	O
and	O
more	O
than	O
22500	O
tests	O
performed	O
.	O
Additionally	O
,	O
a	O
quantisation	O
model	O
was	O
developed	O
to	O
be	O
able	O
to	O
use	O
predicate-based	O
distance	B-KEY
measures	I-KEY
on	O
continuous	O
data	O
as	O
well	O
.	O
The	O
evaluation	O
shows	O
that	O
the	O
distance	B-KEY
measures	I-KEY
recommended	O
in	O
the	O
MPEG-7-standard	O
are	O
among	O
the	O
best	O
but	O
that	O
other	O
measures	O
perform	O
even	O
better	O
.	O

An	O
Architectural	O
Framework	O
and	O
a	O
Middleware	O
for	O
Cooperating	O
Smart	O
Components	O
*	O
U.Lisboa	O
U.Ulm	O
U.Lisboa	O
casim@di.fc.ul.pt	O
kaiser@informatik.uni-	O
pjv@di.fc.ul.pt	O
ulm.de	O
ABSTRACT	O
In	O
a	O
future	O
networked	O
physical	O
world	O
,	O
a	O
myriad	O
of	O
smart	B-KEY
sensors	I-KEY
and	O
actuators	O
assess	O
and	O
control	O
aspects	O
of	O
their	O
environments	O
and	O
autonomously	O
act	O
in	O
response	O
to	O
it	O
.	O
Examples	O
range	O
in	O
telematics	O
,	O
traffic	O
management	O
,	O
team	O
robotics	O
or	O
home	O
automation	O
to	O
name	O
a	O
few	O
.	O
To	O
a	O
large	O
extent	O
,	O
such	O
systems	O
operate	O
proactively	O
and	O
independently	O
of	O
direct	O
human	O
control	O
driven	O
by	O
the	O
perception	O
of	O
the	O
environment	O
and	O
the	O
ability	O
to	O
organize	O
respective	O
computations	O
dynamically	O
.	O
The	O
challenging	O
characteristics	O
of	O
these	O
applications	O
include	O
sentience	O
and	O
autonomy	O
of	O
components	O
,	O
issues	O
of	O
responsiveness	O
and	O
safety	O
criticality	O
,	O
geographical	O
dispersion	O
,	O
mobility	O
and	O
evolution	O
.	O
A	O
crucial	O
design	O
decision	O
is	O
the	O
choice	O
of	O
the	O
appropriate	O
abstractions	O
and	O
interaction	O
mechanisms	O
.	O
Looking	O
to	O
the	O
basic	O
building	O
blocks	O
of	O
such	O
systems	O
we	O
may	O
find	O
components	O
which	O
comprise	O
mechanical	O
components	O
,	O
hardware	O
and	O
software	O
and	O
a	O
network	O
interface	O
,	O
thus	O
these	O
components	O
have	O
different	O
characteristics	O
compared	O
to	O
pure	O
software	O
components	O
.	O
They	O
are	O
able	O
to	O
spontaneously	O
disseminate	O
information	O
in	O
response	O
to	O
events	O
observed	O
in	O
the	O
physical	O
environment	O
or	O
to	O
events	O
received	O
from	O
other	O
component	O
via	O
the	O
network	O
interface	O
.	O
Larger	O
autonomous	O
components	O
may	O
be	O
composed	O
recursively	O
from	O
these	O
building	O
blocks	O
.	O
The	O
paper	O
describes	O
an	O
architectural	O
framework	O
and	O
a	O
middleware	O
supporting	O
a	O
component-based	O
system	O
and	O
an	O
integrated	O
view	O
on	O
events-based	O
communication	O
comprising	O
the	O
real	O
world	O
events	O
and	O
the	O
events	O
generated	O
in	O
the	O
system	O
.	O
It	O
starts	O
by	O
an	O
outline	O
of	O
the	O
component-based	O
system	O
construction	O
.	O
The	O
generic	B-KEY
event	I-KEY
architecture	I-KEY
GEAR	O
is	O
introduced	O
which	O
describes	O
the	O
event-based	O
interaction	O
between	O
the	O
components	O
via	O
a	O
generic	O
event	O
layer	O
.	O
The	O
generic	O
event	O
layer	O
hides	O
the	O
different	O
communication	O
channels	O
including	O
*	O
This	O
work	O
was	O
partially	O
supported	O
by	O
the	O
EC	O
,	O
through	O
project	O
IST-2000-26031	O
-LRB-	O
CORTEX	B-KEY
-RRB-	O
,	O
and	O
by	O
the	O
FCT	O
,	O
through	O
the	O
Large-Scale	O
Informatic	O
Systems	O
Laboratory	O
-LRB-	O
LaSIGE	O
-RRB-	O
and	O
project	O
POSI/1999/CHS	O
/	O
33996	O
-LRB-	O
DEFEATS	O
-RRB-	O
.	O
the	O
interactions	O
through	O
the	O
environment	O
.	O
An	O
appropriate	O
middleware	O
is	O
presented	O
which	O
reflects	O
these	O
needs	O
and	O
allows	O
to	O
specify	O
events	O
which	O
have	O
quality	O
attributes	O
to	O
express	O
temporal	B-KEY
constraints	I-KEY
.	O
This	O
is	O
complemented	O
by	O
the	O
notion	O
of	O
event	B-KEY
channels	I-KEY
which	O
are	O
abstractions	O
of	O
the	O
underlying	O
network	O
and	O
allow	O
to	O
enforce	O
quality	O
attributes	O
.	O
They	O
are	O
established	O
prior	O
to	O
interaction	O
to	O
reserve	O
the	O
needed	O
computational	O
and	O
network	O
resources	O
for	O
highly	O
predictable	O
event	O
dissemination	O
.	O

Context	B-KEY
Awareness	I-KEY
for	O
Group	B-KEY
Interaction	I-KEY
Support	O
ABSTRACT	O
In	O
this	O
paper	O
,	O
we	O
present	O
an	O
implemented	O
system	O
for	O
supporting	O
group	B-KEY
interaction	I-KEY
in	O
mobile	O
distributed	O
computing	O
environments	O
.	O
First	O
,	O
an	O
introduction	O
to	O
context	O
computing	O
and	O
a	O
motivation	O
for	O
using	O
contextual	O
information	O
to	O
facilitate	O
group	B-KEY
interaction	I-KEY
is	O
given	O
.	O
We	O
then	O
present	O
the	O
architecture	O
of	O
our	O
system	O
,	O
which	O
consists	O
of	O
two	O
parts	O
:	O
a	O
subsystem	O
for	O
location	B-KEY
sensing	I-KEY
that	O
acquires	O
information	O
about	O
the	O
location	O
of	O
users	O
as	O
well	O
as	O
spatial	O
proximities	O
between	O
them	O
,	O
and	O
one	O
for	O
the	O
actual	O
context-aware	O
application	O
,	O
which	O
provides	O
services	O
for	O
group	B-KEY
interaction	I-KEY
.	O

Empirical	B-KEY
Mechanism	I-KEY
Design	O
:	O
Methods	O
,	O
with	O
Application	O
to	O
a	O
Supply-Chain	O
Scenario	O
ABSTRACT	O
Our	O
proposed	O
methods	O
employ	O
learning	O
and	O
search	O
techniques	O
to	O
estimate	O
outcome	B-KEY
features	I-KEY
of	I-KEY
interest	I-KEY
as	O
a	O
function	O
of	O
mechanism	O
parameter	B-KEY
settings	I-KEY
.	O
We	O
illustrate	O
our	O
approach	O
with	O
a	O
design	O
task	O
from	O
a	O
supply-chain	O
trading	O
competition	O
.	O
Designers	O
adopted	O
several	O
rule	O
changes	O
in	O
order	O
to	O
deter	O
particular	O
procurement	O
behavior	O
,	O
but	O
the	O
measures	O
proved	O
insufficient	O
.	O
Our	O
empirical	B-KEY
mechanism	I-KEY
analysis	B-KEY
models	O
the	O
relation	O
between	O
a	O
key	O
design	O
parameter	O
and	O
outcomes	O
,	O
confirming	O
the	O
observed	B-KEY
behavior	I-KEY
and	O
indicating	O
that	O
no	O
reasonable	O
parameter	B-KEY
settings	I-KEY
would	O
have	O
been	O
likely	O
to	O
achieve	O
the	O
desired	O
effect	O
.	O
More	O
generally	O
,	O
we	O
show	O
that	O
under	O
certain	O
conditions	O
,	O
the	O
estimator	O
of	O
optimal	O
mechanism	O
parameter	B-KEY
setting	I-KEY
based	O
on	O
empirical	O
data	O
is	O
consistent	O
.	O

Scouts	B-KEY
,	O
Promoters	B-KEY
,	O
and	O
Connectors	B-KEY
:	O
The	O
Roles	O
of	O
Ratings	B-KEY
in	O
Nearest	B-KEY
Neighbor	I-KEY
Collaborative	B-KEY
Filtering	I-KEY
ABSTRACT	O
Recommender	B-KEY
systems	O
aggregate	O
individual	O
user	O
ratings	O
into	O
predictions	O
of	O
products	O
or	O
services	O
that	O
might	O
interest	O
visitors	O
.	O
The	O
quality	O
of	O
this	O
aggregation	B-KEY
process	I-KEY
crucially	O
affects	O
the	O
user	O
experience	O
and	O
hence	O
the	O
effectiveness	O
of	O
recommenders	B-KEY
in	O
e-commerce	O
.	O
We	O
present	O
a	O
novel	O
study	O
that	O
disaggregates	O
global	O
recommender	B-KEY
performance	O
metrics	O
into	O
contributions	O
made	O
by	O
each	O
individual	O
rating	B-KEY
,	O
allowing	O
us	O
to	O
characterize	O
the	O
many	O
roles	O
played	O
by	O
ratings	B-KEY
in	O
nearestneighbor	O
collaborative	B-KEY
filtering	I-KEY
.	O
In	O
particular	O
,	O
we	O
formulate	O
three	O
roles	O
--	O
scouts	B-KEY
,	O
promoters	B-KEY
,	O
and	O
connectors	B-KEY
--	O
that	O
capture	O
how	O
users	O
receive	O
recommendations	B-KEY
,	O
how	O
items	O
get	O
recommended	B-KEY
,	O
and	O
how	O
ratings	B-KEY
of	O
these	O
two	O
types	O
are	O
themselves	O
connected	O
-LRB-	O
resp	O
.	O
-RRB-	O
.	O
These	O
roles	O
find	O
direct	O
uses	O
in	O
improving	O
recommendations	B-KEY
for	O
users	O
,	O
in	O
better	O
targeting	O
of	O
items	O
and	O
,	O
most	O
importantly	O
,	O
in	O
helping	O
monitor	O
the	O
health	O
of	O
the	O
system	O
as	O
a	O
whole	O
.	O
For	O
instance	O
,	O
they	O
can	O
be	O
used	O
to	O
track	O
the	O
evolution	O
of	O
neighborhoods	B-KEY
,	O
to	O
identify	O
rating	B-KEY
subspaces	O
that	O
do	O
not	O
contribute	O
-LRB-	O
or	O
contribute	O
negatively	O
-RRB-	O
to	O
system	O
performance	O
,	O
to	O
enumerate	O
users	O
who	O
are	O
in	O
danger	O
of	O
leaving	O
,	O
and	O
to	O
assess	O
the	O
susceptibility	O
of	O
the	O
system	O
to	O
attacks	O
such	O
as	O
shilling	O
.	O
We	O
argue	O
that	O
the	O
three	O
rating	B-KEY
roles	O
presented	O
here	O
provide	O
broad	O
primitives	O
to	O
manage	O
a	O
recommender	B-KEY
system	O
and	O
its	O
community	O
.	O

Mechanism	B-KEY
Design	I-KEY
for	O
Online	O
Real-Time	O
Scheduling	B-KEY
ABSTRACT	O
For	O
the	O
problem	O
of	O
online	O
real-time	O
scheduling	B-KEY
of	O
jobs	O
on	O
a	O
single	O
processor	O
,	O
previous	O
work	O
presents	O
matching	O
upper	O
and	O
lower	O
bounds	O
on	O
the	O
competitive	B-KEY
ratio	I-KEY
that	O
can	O
be	O
achieved	O
by	O
a	O
deterministic	B-KEY
algorithm	I-KEY
.	O
However	O
,	O
these	O
results	O
only	O
apply	O
to	O
the	O
non-strategic	B-KEY
setting	I-KEY
in	O
which	O
the	O
jobs	O
are	O
released	O
directly	O
to	O
the	O
algorithm	O
.	O
Motivated	O
by	O
emerging	O
areas	O
such	O
as	O
grid	O
computing	O
,	O
we	O
instead	O
consider	O
this	O
problem	O
in	O
an	O
economic	O
setting	O
,	O
in	O
which	O
each	O
job	O
is	O
released	O
to	O
a	O
separate	O
,	O
self-interested	O
agent	O
.	O
The	O
agent	O
can	O
then	O
delay	O
releasing	O
the	O
job	O
to	O
the	O
algorithm	O
,	O
inflate	O
its	O
length	O
,	O
and	O
declare	O
an	O
arbitrary	O
value	O
and	O
deadline	B-KEY
for	O
the	O
job	O
,	O
while	O
the	O
center	O
determines	O
not	O
only	O
the	O
schedule	B-KEY
,	O
but	O
the	O
payment	O
of	O
each	O
agent	O
.	O
For	O
the	O
resulting	O
mechanism	B-KEY
design	I-KEY
problem	O
-LRB-	O
in	O
which	O
we	O
also	O
slightly	O
strengthen	O
an	O
assumption	O
from	O
the	O
non-strategic	B-KEY
setting	I-KEY
-RRB-	O
,	O
we	O
present	O
a	O
mechanism	O
that	O
addresses	O
each	O
incentive	O
issue	O
,	O
while	O
only	O
increasing	O
the	O
competitive	B-KEY
ratio	I-KEY
by	O
one	O
.	O
We	O
then	O
show	O
a	O
matching	O
lower	O
bound	O
for	O
deterministic	B-KEY
mechanisms	I-KEY
that	O
never	O
pay	O
the	O
agents	O
.	O

Context	O
Sensitive	O
Stemming	B-KEY
for	O
Web	B-KEY
Search	I-KEY
ABSTRACT	O
Traditionally	O
,	O
stemming	B-KEY
has	O
been	O
applied	O
to	O
Information	O
Retrieval	O
tasks	O
by	O
transforming	O
words	O
in	O
documents	O
to	O
the	O
their	O
root	O
form	O
before	O
indexing	O
,	O
and	O
applying	O
a	O
similar	O
transformation	O
to	O
query	O
terms	O
.	O
Although	O
it	O
increases	O
recall	O
,	O
this	O
naive	O
strategy	O
does	O
not	O
work	O
well	O
for	O
Web	B-KEY
Search	I-KEY
since	O
it	O
lowers	O
precision	O
and	O
requires	O
a	O
significant	O
amount	O
of	O
additional	O
computation	O
.	O
In	O
this	O
paper	O
,	O
we	O
propose	O
a	O
context	O
sensitive	O
stemming	B-KEY
method	O
that	O
addresses	O
these	O
two	O
issues	O
.	O
Two	O
unique	O
properties	O
make	O
our	O
approach	O
feasible	O
for	O
Web	B-KEY
Search	I-KEY
.	O
First	O
,	O
based	O
on	O
statistical	O
language	B-KEY
modeling	I-KEY
,	O
we	O
perform	O
context	O
sensitive	O
analysis	O
on	O
the	O
query	O
side	O
.	O
We	O
accurately	O
predict	O
which	O
of	O
its	O
morphological	O
variants	O
is	O
useful	O
to	O
expand	O
a	O
query	O
term	O
with	O
before	O
submitting	O
the	O
query	O
to	O
the	O
search	O
engine	O
.	O
This	O
dramatically	O
reduces	O
the	O
number	O
of	O
bad	O
expansions	O
,	O
which	O
in	O
turn	O
reduces	O
the	O
cost	O
of	O
additional	O
computation	O
and	O
improves	O
the	O
precision	O
at	O
the	O
same	O
time	O
.	O
Second	O
,	O
our	O
approach	O
performs	O
a	O
context	B-KEY
sensitive	I-KEY
document	I-KEY
matching	I-KEY
for	O
those	O
expanded	O
variants	O
.	O
This	O
conservative	O
strategy	O
serves	O
as	O
a	O
safeguard	O
against	O
spurious	O
stemming	B-KEY
,	O
and	O
it	O
turns	O
out	O
to	O
be	O
very	O
important	O
for	O
improving	O
precision	O
.	O
Using	O
word	O
pluralization	O
handling	O
as	O
an	O
example	O
of	O
our	O
stemming	B-KEY
approach	O
,	O
our	O
experiments	O
on	O
a	O
major	O
Web	B-KEY
search	I-KEY
engine	O
show	O
that	O
stemming	B-KEY
only	O
29	O
%	O
of	O
the	O
query	O
traffic	O
,	O
we	O
can	O
improve	O
relevance	O
as	O
measured	O
by	O
average	O
Discounted	O
Cumulative	O
Gain	O
-LRB-	O
DCG5	O
-RRB-	O
by	O
6.1	O
%	O
on	O
these	O
queries	O
and	O
1.8	O
%	O
over	O
all	O
query	O
traffic	O
.	O

Self-interested	O
Automated	B-KEY
Mechanism	I-KEY
Design	I-KEY
and	O
Implications	O
for	O
Optimal	O
Combinatorial	O
Auctions	O
∗	O
ABSTRACT	O
Often	O
,	O
an	O
outcome	O
must	O
be	O
chosen	O
on	O
the	O
basis	O
of	O
the	O
preferences	O
reported	O
by	O
a	O
group	O
of	O
agents	O
.	O
The	O
key	O
difficulty	O
is	O
that	O
the	O
agents	O
may	O
report	O
their	O
preferences	O
insincerely	O
to	O
make	O
the	O
chosen	O
outcome	O
more	O
favorable	O
to	O
themselves	O
.	O
Mechanism	B-KEY
design	I-KEY
is	O
the	O
art	O
of	O
designing	O
the	O
rules	O
of	O
the	O
game	O
so	O
that	O
the	O
agents	O
are	O
motivated	O
to	O
report	O
their	O
preferences	O
truthfully	O
,	O
and	O
a	O
desirable	B-KEY
outcome	I-KEY
is	O
chosen	O
.	O
In	O
a	O
recently	O
proposed	O
approach	O
--	O
called	O
automated	B-KEY
mechanism	I-KEY
design	I-KEY
--	O
a	O
mechanism	O
is	O
computed	O
for	O
the	O
preference	O
aggregation	O
setting	O
at	O
hand	O
.	O
This	O
has	O
several	O
advantages	O
,	O
but	O
the	O
downside	O
is	O
that	O
the	O
mechanism	B-KEY
design	I-KEY
optimization	O
problem	O
needs	O
to	O
be	O
solved	O
anew	O
each	O
time	O
.	O
Unlike	O
the	O
earlier	O
work	O
on	O
automated	B-KEY
mechanism	I-KEY
design	I-KEY
that	O
studied	O
a	O
benevolent	O
designer	O
,	O
in	O
this	O
paper	O
we	O
study	O
automated	O
mechanism	O
design	O
problems	O
where	O
the	O
designer	O
is	O
self-interested	O
.	O
In	O
this	O
case	O
,	O
the	O
center	O
cares	O
only	O
about	O
which	O
outcome	O
is	O
chosen	O
and	O
what	O
payments	O
are	O
made	O
to	O
it	O
.	O
The	O
reason	O
that	O
the	O
agents	O
'	O
preferences	O
are	O
relevant	O
is	O
that	O
the	O
center	O
is	O
constrained	O
to	O
making	O
each	O
agent	O
at	O
least	O
as	O
well	O
off	O
as	O
the	O
agent	O
would	O
have	O
been	O
had	O
it	O
not	O
participated	O
in	O
the	O
mechanism	O
.	O
In	O
this	O
setting	O
,	O
we	O
show	O
that	O
designing	O
optimal	O
deterministic	O
mechanisms	O
is	O
NP-complete	O
in	O
two	O
important	O
special	O
cases	O
:	O
when	O
the	O
center	O
is	O
interested	O
only	O
in	O
the	O
payments	O
made	O
to	O
it	O
,	O
and	O
when	O
payments	O
are	O
not	O
possible	O
and	O
the	O
center	O
is	O
interested	O
only	O
in	O
the	O
outcome	O
chosen	O
.	O
We	O
then	O
show	O
how	O
allowing	O
for	O
randomization	O
in	O
the	O
mechanism	O
makes	O
problems	O
in	O
this	O
setting	O
computationally	O
easy	O
.	O
Finally	O
,	O
we	O
show	O
that	O
the	O
payment-maximizing	O
AMD	O
problem	O
is	O
closely	O
related	O
to	O
an	O
interesting	O
variant	O
of	O
the	O
optimal	O
-LRB-	O
revenuemaximizing	O
-RRB-	O
combinatorial	B-KEY
auction	I-KEY
design	O
problem	O
,	O
where	O
the	O
bidders	O
have	O
``	O
best-only	O
''	O
preferences	O
.	O
We	O
show	O
that	O
here	O
,	O
too	O
,	O
designing	O
an	O
optimal	O
deterministic	O
auction	O
is	O
NPcomplete	O
,	O
but	O
designing	O
an	O
optimal	O
randomized	O
auction	O
is	O
easy	O
.	O
∗	O
Supported	O
by	O
NSF	O
under	O
CAREER	O
Award	O
IRI-9703122	O
,	O
Grant	O
IIS-9800994	O
,	O
ITR	O
IIS-0081246	O
,	O
and	O
ITR	O
IIS-0121678	O
.	O

Searching	O
for	O
Joint	O
Gains	O
in	O
Automated	B-KEY
Negotiations	I-KEY
Based	O
on	O
Multi-criteria	O
Decision	O
Making	O
Theory	O
ABSTRACT	O
It	O
is	O
well	O
established	O
by	O
conflict	O
theorists	O
and	O
others	O
that	O
successful	O
negotiation	B-KEY
should	O
incorporate	O
``	O
creating	B-KEY
value	I-KEY
''	O
as	O
well	O
as	O
``	O
claiming	B-KEY
value	I-KEY
.	O
''	O
Joint	O
improvements	O
that	O
bring	O
benefits	O
to	O
all	O
parties	O
can	O
be	O
realised	O
by	O
-LRB-	O
i	O
-RRB-	O
identifying	O
attributes	O
that	O
are	O
not	O
of	O
direct	O
conflict	O
between	O
the	O
parties	O
,	O
-LRB-	O
ii	O
-RRB-	O
tradeoffs	O
on	O
attributes	O
that	O
are	O
valued	O
differently	O
by	O
different	O
parties	O
,	O
and	O
-LRB-	O
iii	O
-RRB-	O
searching	O
for	O
values	O
within	O
attributes	O
that	O
could	O
bring	O
more	O
gains	O
to	O
one	O
party	O
while	O
not	O
incurring	O
too	O
much	O
loss	O
on	O
the	O
other	O
party	O
.	O
In	O
this	O
paper	O
we	O
propose	O
an	O
approach	O
for	O
maximising	O
joint	O
gains	O
in	O
automated	B-KEY
negotiations	I-KEY
by	O
formulating	O
the	O
negotiation	O
problem	O
as	O
a	O
multi-criteria	O
decision	O
making	O
problem	O
and	O
taking	O
advantage	O
of	O
several	O
optimisation	O
techniques	O
introduced	O
by	O
operations	O
researchers	O
and	O
conflict	O
theorists	O
.	O
We	O
use	O
a	O
mediator	B-KEY
to	O
protect	O
the	O
negotiating	B-KEY
parties	O
from	O
unnecessary	O
disclosure	O
of	O
information	O
to	O
their	O
opponent	O
,	O
while	O
also	O
allowing	O
an	O
objective	O
calculation	O
of	O
maximum	O
joint	O
gains	O
.	O
We	O
separate	O
out	O
attributes	O
that	O
take	O
a	O
finite	O
set	O
of	O
values	O
-LRB-	O
simple	O
attributes	O
-RRB-	O
from	O
those	O
with	O
continuous	O
values	O
,	O
and	O
we	O
show	O
that	O
for	O
simple	O
attributes	O
,	O
the	O
mediator	B-KEY
can	O
determine	O
the	O
Pareto-optimal	O
values	O
.	O
In	O
addition	O
we	O
show	O
that	O
if	O
none	O
of	O
the	O
simple	O
attributes	O
strongly	O
dominates	O
the	O
other	O
simple	O
attributes	O
,	O
then	O
truth	O
telling	O
is	O
an	O
equilibrium	O
strategy	O
for	O
negotiators	B-KEY
during	O
the	O
optimisation	O
of	O
simple	O
attributes	O
.	O
We	O
also	O
describe	O
an	O
approach	O
for	O
improving	O
joint	O
gains	O
on	O
non-simple	O
attributes	O
,	O
by	O
moving	O
the	O
parties	O
in	O
a	O
series	O
of	O
steps	O
,	O
towards	O
the	O
Pareto-optimal	O
frontier	O
.	O

Shooter	O
Localization	O
and	O
Weapon	B-KEY
Classification	I-KEY
with	O
Soldier-Wearable	O
Networked	O
Sensors	O
ABSTRACT	O
The	O
paper	O
presents	O
a	O
wireless	O
sensor	O
network-based	O
mobile	O
countersniper	O
system	O
.	O
A	O
sensor	O
node	O
consists	O
of	O
a	O
helmetmounted	O
microphone	O
array	O
,	O
a	O
COTS	O
MICAz	O
mote	O
for	O
internode	B-KEY
communication	I-KEY
and	O
a	O
custom	O
sensorboard	B-KEY
that	O
implements	O
the	O
acoustic	O
detection	O
and	O
Time	O
of	O
Arrival	O
-LRB-	O
ToA	O
-RRB-	O
estimation	O
algorithms	O
on	O
an	O
FPGA	O
.	O
A	O
3-axis	O
compass	O
provides	O
self	B-KEY
orientation	I-KEY
and	O
Bluetooth	O
is	O
used	O
for	O
communication	O
with	O
the	O
soldier	O
's	O
PDA	O
running	O
the	O
data	B-KEY
fusion	I-KEY
and	O
the	O
user	O
interface	O
.	O
The	O
heterogeneous	O
sensor	O
fusion	O
algorithm	O
can	O
work	O
with	O
data	O
from	O
a	O
single	O
sensor	O
or	O
it	O
can	O
fuse	O
ToA	O
or	O
Angle	O
of	O
Arrival	O
-LRB-	O
AoA	O
-RRB-	O
observations	O
of	O
muzzle	O
blasts	O
and	O
ballistic	O
shockwaves	O
from	O
multiple	O
sensors	O
.	O
The	O
system	O
estimates	O
the	O
trajectory	B-KEY
,	O
the	O
range	B-KEY
,	O
the	O
caliber	B-KEY
and	O
the	O
weapon	B-KEY
type	I-KEY
.	O
The	O
paper	O
presents	O
the	O
system	O
design	O
and	O
the	O
results	O
from	O
an	O
independent	O
evaluation	O
at	O
the	O
US	O
Army	O
Aberdeen	O
Test	O
Center	O
.	O
The	O
system	O
performance	O
is	O
characterized	O
by	O
1degree	O
trajectory	B-KEY
precision	O
and	O
over	O
95	O
%	O
caliber	O
estimation	O
accuracy	O
for	O
all	O
shots	O
,	O
and	O
close	O
to	O
100	O
%	O
weapon	O
estimation	O
accuracy	O
for	O
4	O
out	O
of	O
6	O
guns	O
tested	O
.	O

A	O
Multilateral	O
Multi-issue	B-KEY
Negotiation	I-KEY
Protocol	O
ABSTRACT	O
In	O
this	O
paper	O
,	O
we	O
present	O
a	O
new	O
protocol	O
to	O
address	O
multilateral	O
multi-issue	B-KEY
negotiation	I-KEY
in	O
a	O
cooperative	O
context	O
.	O
We	O
consider	O
complex	O
dependencies	O
between	O
multiple	O
issues	O
by	O
modelling	B-KEY
the	O
preferences	O
of	O
the	O
agents	O
with	O
a	O
multi-criteria	O
decision	O
aid	O
tool	O
,	O
also	O
enabling	O
us	O
to	O
extract	O
relevant	O
information	O
on	O
a	O
proposal	O
assessment	O
.	O
This	O
information	O
is	O
used	O
in	O
the	O
protocol	O
to	O
help	O
in	O
accelerating	O
the	O
search	O
for	O
a	O
consensus	O
between	O
the	O
cooperative	B-KEY
agents	I-KEY
.	O
In	O
addition	O
,	O
the	O
negotiation	O
procedure	O
is	O
defined	O
in	O
a	O
crisis	B-KEY
management	I-KEY
context	O
where	O
the	O
common	O
objective	O
of	O
our	O
agents	O
is	O
also	O
considered	O
in	O
the	O
preferences	O
of	O
a	O
mediator	O
agent	O
.	O

Authority	B-KEY
Assignment	O
in	O
Distributed	O
Multi-Player	O
Proxy-based	O
Games	O
ABSTRACT	O
We	O
present	O
a	O
proxy-based	O
gaming	O
architecture	O
and	O
authority	B-KEY
assignment	O
within	O
this	O
architecture	O
that	O
can	O
lead	O
to	O
better	O
game	O
playing	O
experience	O
in	O
Massively	O
Multi-player	O
Online	O
games	O
.	O
The	O
proposed	O
game	O
architecture	O
consists	O
of	O
distributed	O
game	O
clients	O
that	O
connect	O
to	O
game	O
proxies	O
-LRB-	O
referred	O
to	O
as	O
``	O
communication	B-KEY
proxies	I-KEY
''	O
-RRB-	O
which	O
forward	O
game	O
related	O
messages	O
from	O
the	O
clients	O
to	O
one	O
or	O
more	O
game	O
servers	O
.	O
Unlike	O
proxy-based	O
architectures	O
that	O
have	O
been	O
proposed	O
in	O
the	O
literature	O
where	O
the	O
proxies	O
replicate	O
all	O
of	O
the	O
game	O
state	O
,	O
the	O
communication	B-KEY
proxies	I-KEY
in	O
the	O
proposed	O
architecture	O
support	O
clients	O
that	O
are	O
in	O
proximity	O
to	O
it	O
in	O
the	O
physical	O
network	O
and	O
maintain	O
information	O
about	O
selected	O
portions	O
of	O
the	O
game	O
space	O
that	O
are	O
relevant	O
only	O
to	O
the	O
clients	O
that	O
they	O
support	O
.	O
Using	O
this	O
architecture	O
,	O
we	O
propose	O
an	O
authority	B-KEY
assignment	O
mechanism	O
that	O
divides	O
the	O
authority	O
for	O
deciding	O
the	O
outcome	O
of	O
different	O
actions/events	O
that	O
occur	O
within	O
the	O
game	O
between	O
client	O
and	O
servers	O
on	O
a	O
per	O
action/event	O
basis	O
.	O
We	O
show	O
that	O
such	O
division	O
of	O
authority	B-KEY
leads	O
to	O
a	O
smoother	O
game	O
playing	O
experience	O
by	O
implementing	O
this	O
mechanism	O
in	O
a	O
massively	O
multi-player	O
online	O
game	O
called	O
RPGQuest	O
.	O
In	O
addition	O
,	O
we	O
argue	O
that	O
cheat	O
detection	O
techniques	O
can	O
be	O
easily	O
implemented	O
at	O
the	O
communication	B-KEY
proxies	I-KEY
if	O
they	O
are	O
made	O
aware	O
of	O
the	O
game-play	O
mechanics	O
.	O

An	O
Evaluation	O
of	O
Availability	B-KEY
Latency	I-KEY
in	O
Carrier-based	O
Vehicular	O
Ad-Hoc	O
Networks	O
Shahram	O
ABSTRACT	O
On-demand	O
delivery	O
of	O
audio	B-KEY
and	I-KEY
video	I-KEY
clips	I-KEY
in	O
peer-to-peer	O
vehicular	O
ad-hoc	O
networks	O
is	O
an	O
emerging	O
area	O
of	O
research	O
.	O
Our	O
target	O
environment	O
uses	O
data	B-KEY
carriers	I-KEY
,	O
termed	B-KEY
zebroids	I-KEY
,	O
where	O
a	O
mobile	O
device	O
carries	O
a	O
data	O
item	O
on	O
behalf	O
of	O
a	O
server	O
to	O
a	O
client	O
thereby	O
minimizing	O
its	O
availability	O
latency	O
.	O
In	O
this	O
study	O
,	O
we	O
quantify	O
the	O
variation	O
in	O
availability	B-KEY
latency	I-KEY
with	O
zebroids	O
as	O
a	O
function	O
of	O
a	O
rich	O
set	O
of	O
parameters	O
such	O
as	O
car	O
density	O
,	O
storage	O
per	O
device	O
,	O
repository	O
size	O
,	O
and	O
replacement	O
policies	O
employed	O
by	O
zebroids	O
.	O
Using	O
analysis	O
and	O
extensive	O
simulations	O
,	O
we	O
gain	O
novel	O
insights	O
into	O
the	O
design	O
of	O
carrier-based	O
systems	O
.	O
Significant	O
improvements	O
in	O
latency	B-KEY
can	O
be	O
obtained	O
with	O
zebroids	B-KEY
at	O
the	O
cost	O
of	O
a	O
minimal	O
overhead	O
.	O
These	O
improvements	O
occur	O
even	O
in	O
scenarios	O
with	O
lower	O
accuracy	O
in	O
the	O
predictions	O
of	O
the	O
car	O
routes	O
.	O
Two	O
particularly	O
surprising	O
findings	O
are	O
:	O
-LRB-	O
1	O
-RRB-	O
a	O
naive	B-KEY
random	I-KEY
replacement	I-KEY
policy	I-KEY
employed	O
by	O
the	O
zebroids	O
shows	O
competitive	O
performance	O
,	O
and	O
-LRB-	O
2	O
-RRB-	O
latency	O
improvements	O
obtained	O
with	O
a	O
simplified	O
instantiation	O
of	O
zebroids	O
are	O
found	O
to	O
be	O
robust	O
to	O
changes	O
in	O
the	O
popularity	O
distribution	O
of	O
the	O
data	O
items	O
.	O
Categories	O
and	O
Subject	O
Descriptors	O
:	O
C.	O
2.4	O
-LSB-	O
Distributed	O
Systems	O
-RSB-	O
:	O
Client/Server	O

From	O
Optimal	O
Limited	O
To	O
Unlimited	B-KEY
Supply	I-KEY
Auctions	B-KEY
ABSTRACT	O
We	O
investigate	O
the	O
class	O
of	O
single-round	O
,	O
sealed-bid	O
auctions	B-KEY
for	O
a	O
set	O
of	O
identical	O
items	O
to	O
bidders	O
who	O
each	O
desire	O
one	O
unit	O
.	O
We	O
adopt	O
the	O
worst-case	O
competitive	O
framework	O
defined	O
by	O
-LSB-	O
9	O
,	O
5	O
-RSB-	O
that	O
compares	O
the	O
profit	O
of	O
an	O
auction	B-KEY
to	O
that	O
of	O
an	O
optimal	O
single-price	O
sale	O
of	O
least	O
two	O
items	O
.	O
In	O
this	O
paper	O
,	O
we	O
first	O
derive	O
an	O
optimal	O
auction	B-KEY
for	O
three	O
items	O
,	O
answering	O
an	O
open	O
question	O
from	O
-LSB-	O
8	O
-RSB-	O
.	O
Second	O
,	O
we	O
show	O
that	O
the	O
form	O
of	O
this	O
auction	B-KEY
is	O
independent	O
of	O
the	O
competitive	O
framework	O
used	O
.	O
Third	O
,	O
we	O
propose	O
a	O
schema	O
for	O
converting	O
a	O
given	O
limited-supply	O
auction	B-KEY
into	O
an	O
unlimited	B-KEY
supply	I-KEY
auction	B-KEY
.	O
Applying	O
this	O
technique	O
to	O
our	O
optimal	O
auction	B-KEY
for	O
three	O
items	O
,	O
we	O
achieve	O
an	O
auction	B-KEY
with	O
a	O
competitive	O
ratio	B-KEY
of	O
3.25	O
,	O
which	O
improves	O
upon	O
the	O
previously	O
best-known	O
competitive	O
ratio	B-KEY
of	O
3.39	O
from	O
-LSB-	O
7	O
-RSB-	O
.	O
Finally	O
,	O
we	O
generalize	O
a	O
result	O
from	O
-LSB-	O
8	O
-RSB-	O
and	O
extend	O
our	O
understanding	O
of	O
the	O
nature	O
of	O
the	O
optimal	O
competitive	O
auction	B-KEY
by	O
showing	O
that	O
the	O
optimal	O
competitive	O
auction	B-KEY
occasionally	O
offers	O
prices	O
that	O
are	O
higher	O
than	O
all	O
bid	O
values	O
.	O

A	O
Holistic	O
Approach	O
to	O
High-Performance	O
Computing	O
:	O
Xgrid	B-KEY
Experience	O
ABSTRACT	O
The	O
Ringling	O
School	O
of	O
Art	O
and	O
Design	B-KEY
is	O
a	O
fully	O
accredited	O
fouryear	O
college	O
of	O
visual	B-KEY
arts	I-KEY
and	O
design	B-KEY
.	O
With	O
a	O
student	O
to	O
computer	O
ratio	O
of	O
better	O
than	O
2-to-1	O
,	O
the	O
Ringling	O
School	O
has	O
achieved	O
national	O
recognition	O
for	O
its	O
large-scale	O
integration	O
of	O
technology	O
into	O
collegiate	O
visual	B-KEY
art	I-KEY
and	O
design	B-KEY
education	O
.	O
We	O
have	O
found	O
that	O
Mac	B-KEY
OS	I-KEY
X	I-KEY
is	O
the	O
best	O
operating	B-KEY
system	I-KEY
to	O
train	O
future	O
artists	O
and	O
designers	B-KEY
.	O
Moreover	O
,	O
we	O
can	O
now	O
buy	O
Macs	O
to	O
run	O
high-end	B-KEY
graphics	I-KEY
,	O
nonlinear	B-KEY
video	I-KEY
editing	I-KEY
,	O
animation	B-KEY
,	O
multimedia	B-KEY
,	O
web	B-KEY
production	I-KEY
,	O
and	O
digital	B-KEY
video	I-KEY
applications	I-KEY
rather	O
than	O
expensive	O
UNIX	O
workstations	O
.	O
As	O
visual	O
artists	O
cross	O
from	O
paint	O
on	O
canvas	O
to	O
creating	O
in	O
the	O
digital	O
realm	O
,	O
the	O
demand	O
for	O
a	O
highperformance	B-KEY
computing	I-KEY
environment	O
grows	O
.	O
In	O
our	O
public	O
computer	O
laboratories	O
,	O
students	O
use	O
the	O
computers	O
most	O
often	O
during	O
the	O
workday	O
;	O
at	O
night	O
and	O
on	O
weekends	O
the	O
computers	O
see	O
only	O
light	O
use	O
.	O
In	O
order	O
to	O
harness	O
the	O
lost	O
processing	O
time	O
for	O
tasks	O
such	O
as	O
video	O
rendering	B-KEY
,	O
we	O
are	O
testing	O
Xgrid	B-KEY
,	O
a	O
suite	O
of	O
Mac	B-KEY
OS	I-KEY
X	I-KEY
applications	O
recently	O
developed	O
by	O
Apple	O
for	O
parallel	O
and	O
distributed	O
high-performance	O
computing	O
.	O
As	O
with	O
any	O
new	O
technology	O
deployment	O
,	O
IT	O
managers	O
need	O
to	O
consider	O
a	O
number	O
of	O
factors	O
as	O
they	O
assess	O
,	O
plan	O
,	O
and	O
implement	O
Xgrid	B-KEY
.	O
Therefore	O
,	O
we	O
would	O
like	O
to	O
share	O
valuable	O
information	O
we	O
learned	O
from	O
our	O
implementation	O
of	O
an	O
Xgrid	B-KEY
environment	O
with	O
our	O
colleagues	O
.	O
In	O
our	O
report	O
,	O
we	O
will	O
address	O
issues	O
such	O
as	O
assessing	O
the	O
needs	O
for	O
grid	B-KEY
computing	I-KEY
,	O
potential	O
applications	O
,	O
management	O
tools	O
,	O
security	O
,	O
authentication	O
,	O
integration	O
into	O
existing	O
infrastructure	O
,	O
application	O
support	O
,	O
user	O
training	O
,	O
and	O
user	O
support	O
.	O
Furthermore	O
,	O
we	O
will	O
discuss	O
the	O
issues	O
that	O
arose	O
and	O
the	O
lessons	O
learned	O
during	O
and	O
after	O
the	O
implementation	O
process	O
.	O

A	O
Q-decomposition	B-KEY
and	O
Bounded	O
RTDP	O
Approach	O
to	O
Resource	B-KEY
Allocation	I-KEY
ABSTRACT	O
This	O
paper	O
contributes	O
to	O
solve	O
effectively	O
stochastic	O
resource	B-KEY
allocation	I-KEY
problems	O
known	O
to	O
be	O
NP-Complete	O
.	O
To	O
address	O
this	O
complex	O
resource	B-KEY
management	I-KEY
problem	O
,	O
a	O
Qdecomposition	O
approach	O
is	O
proposed	O
when	O
the	O
resources	O
which	O
are	O
already	O
shared	O
among	O
the	O
agents	O
,	O
but	O
the	O
actions	O
made	O
by	O
an	O
agent	O
may	O
influence	O
the	O
reward	O
obtained	O
by	O
at	O
least	O
another	O
agent	O
.	O
The	O
Q-decomposition	B-KEY
allows	O
to	O
coordinate	O
these	O
reward	B-KEY
separated	I-KEY
agents	I-KEY
and	O
thus	O
permits	O
to	O
reduce	O
the	O
set	O
of	O
states	O
and	O
actions	O
to	O
consider	O
.	O
On	O
the	O
other	O
hand	O
,	O
when	O
the	O
resources	O
are	O
available	O
to	O
all	O
agents	O
,	O
no	O
Qdecomposition	O
is	O
possible	O
and	O
we	O
use	O
heuristic	B-KEY
search	I-KEY
.	O
In	O
particular	O
,	O
the	O
bounded	O
Real-time	O
Dynamic	O
Programming	O
-LRB-	O
bounded	O
RTDP	O
-RRB-	O
is	O
used	O
.	O
Bounded	O
RTDP	O
concentrates	O
the	O
planning	O
on	O
significant	O
states	O
only	O
and	O
prunes	O
the	O
action	O
space	O
.	O
The	O
pruning	O
is	O
accomplished	O
by	O
proposing	O
tight	O
upper	O
and	O
lower	O
bounds	O
on	O
the	O
value	O
function	O
.	O

Beyond	O
PageRank	B-KEY
:	O
Machine	B-KEY
Learning	I-KEY
for	O
Static	B-KEY
Ranking	I-KEY
ABSTRACT	O
Since	O
the	O
publication	O
of	O
Brin	O
and	O
Page	O
's	O
paper	O
on	O
PageRank	B-KEY
,	O
many	O
in	O
the	O
Web	O
community	O
have	O
depended	O
on	O
PageRank	B-KEY
for	O
the	O
static	O
-LRB-	O
query-independent	O
-RRB-	O
ordering	O
of	O
Web	O
pages	O
.	O
We	O
show	O
that	O
we	O
can	O
significantly	O
outperform	O
PageRank	B-KEY
using	O
features	O
that	O
are	O
independent	O
of	O
the	O
link	O
structure	O
of	O
the	O
Web	O
.	O
We	O
gain	O
a	O
further	O
boost	O
in	O
accuracy	O
by	O
using	O
data	O
on	O
the	O
frequency	O
at	O
which	O
users	O
visit	O
Web	O
pages	O
.	O
We	O
use	O
RankNet	B-KEY
,	O
a	O
ranking	O
machine	B-KEY
learning	I-KEY
algorithm	O
,	O
to	O
combine	O
these	O
and	O
other	O
static	O
features	O
based	O
on	O
anchor	O
text	O
and	O
domain	O
characteristics	O
.	O
The	O
resulting	O
model	O
achieves	O
a	O
static	B-KEY
ranking	I-KEY
pairwise	O
accuracy	O
of	O
67.3	O
%	O
-LRB-	O
vs.	O
56.7	O
%	O
for	O
PageRank	B-KEY
or	O
50	O
%	O
for	O
random	O
-RRB-	O
.	O

Impedance	O
Coupling	O
in	O
Content-targeted	B-KEY
Advertising	I-KEY
ABSTRACT	O
The	O
current	O
boom	O
of	O
the	O
Web	B-KEY
is	O
associated	O
with	O
the	O
revenues	O
originated	O
from	O
on-line	O
advertising	B-KEY
.	O
While	O
search-based	O
advertising	B-KEY
is	O
dominant	O
,	O
the	O
association	O
of	O
ads	O
with	O
a	O
Web	B-KEY
page	O
-LRB-	O
during	O
user	O
navigation	O
-RRB-	O
is	O
becoming	O
increasingly	O
important	O
.	O
In	O
this	O
work	O
,	O
we	O
study	O
the	O
problem	O
of	O
associating	O
ads	O
with	O
a	O
Web	B-KEY
page	O
,	O
referred	O
to	O
as	O
content-targeted	B-KEY
advertising	I-KEY
,	O
from	O
a	O
computer	O
science	O
perspective	O
.	O
We	O
assume	O
that	O
we	O
have	O
access	O
to	O
the	O
text	O
of	O
the	O
Web	B-KEY
page	O
,	O
the	O
keywords	O
declared	O
by	O
an	O
advertiser	B-KEY
,	O
and	O
a	O
text	O
associated	O
with	O
the	O
advertiser	B-KEY
's	O
business	O
.	O
Using	O
no	O
other	O
information	O
and	O
operating	O
in	O
fully	O
automatic	O
fashion	O
,	O
we	O
propose	O
ten	O
strategies	O
for	O
solving	O
the	O
problem	O
and	O
evaluate	O
their	O
effectiveness	O
.	O
Our	O
methods	O
indicate	O
that	O
a	O
matching	B-KEY
strategy	I-KEY
that	O
takes	O
into	O
account	O
the	O
semantics	O
of	O
the	O
problem	O
-LRB-	O
referred	O
to	O
as	O
AAK	O
for	O
``	O
ads	B-KEY
and	I-KEY
keywords	I-KEY
''	O
-RRB-	O
can	O
yield	O
gains	O
in	O
average	O
precision	O
figures	O
of	O
60	O
%	O
compared	O
to	O
a	O
trivial	O
vector-based	O
strategy	O
.	O
Further	O
,	O
a	O
more	O
sophisticated	O
impedance	B-KEY
coupling	I-KEY
strategy	I-KEY
,	O
which	O
expands	O
the	O
text	O
of	O
the	O
Web	B-KEY
page	O
to	O
reduce	O
vocabulary	O
impedance	O
with	O
regard	O
to	O
an	O
advertisement	B-KEY
,	O
can	O
yield	O
extra	O
gains	O
in	O
average	O
precision	O
of	O
50	O
%	O
.	O
These	O
are	O
first	O
results	O
.	O
They	O
suggest	O
that	O
great	O
accuracy	O
in	O
content-targeted	B-KEY
advertising	I-KEY
can	O
be	O
attained	O
with	O
appropriate	O
algorithms	O
.	O

SIGIR	O
2007	O
Proceedings	O
Session	O
20	O
:	O
Link	O
Analysis	O
HITS	B-KEY
on	O
the	O
Web	O
:	O
How	O
does	O
it	O
Compare	O
?	O
*	O
ABSTRACT	O
This	O
paper	O
describes	O
a	O
large-scale	O
evaluation	O
of	O
the	O
effectiveness	O
of	O
HITS	B-KEY
in	O
comparison	O
with	O
other	O
link-based	O
ranking	B-KEY
algorithms	O
,	O
when	O
used	O
in	O
combination	O
with	O
a	O
state-ofthe-art	O
text	O
retrieval	O
algorithm	O
exploiting	O
anchor	O
text	O
.	O
We	O
quantified	O
their	O
effectiveness	O
using	O
three	O
common	O
performance	O
measures	O
:	O
the	O
mean	B-KEY
reciprocal	I-KEY
rank	I-KEY
,	O
the	O
mean	O
average	O
precision	O
,	O
and	O
the	O
normalized	O
discounted	O
cumulative	O
gain	O
measurements	O
.	O
The	O
evaluation	O
is	O
based	O
on	O
two	O
large	O
data	O
sets	O
:	O
a	O
breadth-first	B-KEY
search	I-KEY
crawl	I-KEY
of	O
463	O
million	O
web	O
pages	O
containing	O
17.6	O
billion	O
hyperlinks	O
and	O
referencing	O
2.9	O
billion	O
distinct	O
URLs	O
;	O
and	O
a	O
set	O
of	O
28,043	O
queries	O
sampled	O
from	O
a	O
query	O
log	O
,	O
each	O
query	O
having	O
on	O
average	O
2,383	O
results	O
,	O
about	O
17	O
of	O
which	O
were	O
labeled	O
by	O
judges	O
.	O
We	O
found	O
that	O
HITS	B-KEY
outperforms	O
PageRank	B-KEY
,	O
but	O
is	O
about	O
as	O
effective	O
as	O
web-page	O
in-degree	O
.	O
The	O
same	O
holds	O
true	O
when	O
any	O
of	O
the	O
link-based	O
features	O
are	O
combined	O
with	O
the	O
text	O
retrieval	O
algorithm	O
.	O
Finally	O
,	O
we	O
studied	O
the	O
relationship	O
between	O
query	B-KEY
specificity	I-KEY
and	O
the	O
effectiveness	O
of	O
selected	O
features	O
,	O
and	O
found	O
that	O
link-based	O
features	O
perform	O
better	O
for	O
general	O
queries	O
,	O
whereas	O
BM25F	B-KEY
performs	O
better	O
for	O
specific	O
queries	O
.	O

An	O
Analysis	O
of	O
Alternative	B-KEY
Slot	I-KEY
Auction	I-KEY
Designs	I-KEY
for	O
Sponsored	B-KEY
Search	I-KEY
ABSTRACT	O
Billions	O
of	O
dollars	O
are	O
spent	O
each	O
year	O
on	O
sponsored	B-KEY
search	I-KEY
,	O
a	O
form	O
of	O
advertising	O
where	O
merchants	O
pay	O
for	O
placement	O
alongside	O
web	O
search	O
results	O
.	O
Slots	O
for	O
ad	B-KEY
listings	I-KEY
are	O
allocated	O
via	O
an	O
auction-style	O
mechanism	O
where	O
the	O
higher	O
a	O
merchant	O
bids	O
,	O
the	O
more	O
likely	O
his	O
ad	O
is	O
to	O
appear	O
above	O
other	O
ads	O
on	O
the	O
page	O
.	O
In	O
this	O
paper	O
we	O
analyze	O
the	O
incentive	O
,	O
efficiency	O
,	O
and	O
revenue	O
properties	O
of	O
two	O
slot	O
auction	O
designs	O
:	O
``	O
rank	B-KEY
by	I-KEY
bid	I-KEY
''	O
-LRB-	O
RBB	O
-RRB-	O
and	O
``	O
rank	B-KEY
by	I-KEY
revenue	I-KEY
''	O
-LRB-	O
RBR	O
-RRB-	O
,	O
which	O
correspond	O
to	O
stylized	O
versions	O
of	O
the	O
mechanisms	O
currently	O
used	O
by	O
Yahoo!	O
and	O
Google	O
,	O
respectively	O
.	O
We	O
also	O
consider	O
first	O
-	O
and	O
second-price	O
payment	O
rules	O
together	O
with	O
each	O
of	O
these	O
allocation	O
rules	O
,	O
as	O
both	O
have	O
been	O
used	O
historically	O
.	O
We	O
consider	O
both	O
the	O
``	O
short-run	O
''	O
incomplete	B-KEY
information	I-KEY
setting	O
and	O
the	O
``	O
long-run	O
''	O
complete	O
information	O
setting	O
.	O
With	O
incomplete	B-KEY
information	I-KEY
,	O
neither	O
RBB	O
nor	O
RBR	O
are	O
truthful	O
with	O
either	O
first	O
or	O
second	B-KEY
pricing	I-KEY
.	O
We	O
find	O
that	O
the	O
informational	O
requirements	O
of	O
RBB	O
are	O
much	O
weaker	O
than	O
those	O
of	O
RBR	O
,	O
but	O
that	O
RBR	O
is	O
efficient	O
whereas	O
RBB	O
is	O
not	O
.	O
We	O
also	O
show	O
that	O
no	O
revenue	O
ranking	O
of	O
RBB	O
and	O
RBR	O
is	O
possible	O
given	O
an	O
arbitrary	O
distribution	O
over	O
bidder	O
values	O
and	O
relevance	O
.	O
With	O
complete	O
information	O
,	O
we	O
find	O
that	O
no	O
equilibrium	O
exists	O
with	O
first	O
pricing	O
using	O
either	O
RBB	O
or	O
RBR	O
.	O
We	O
show	O
that	O
there	O
typically	O
exists	O
a	O
multitude	O
of	O
equilibria	O
with	O
second	B-KEY
pricing	I-KEY
,	O
and	O
we	O
bound	O
the	O
divergence	O
of	O
-LRB-	O
economic	O
-RRB-	O
value	O
in	O
such	O
equilibria	O
from	O
the	O
value	O
obtained	O
assuming	O
all	O
merchants	O
bid	O
truthfully	O
.	O

Broad	B-KEY
Expertise	I-KEY
Retrieval	I-KEY
in	O
Sparse	B-KEY
Data	I-KEY
Environments	I-KEY
ABSTRACT	O
Expertise	O
retrieval	O
has	O
been	O
largely	O
unexplored	O
on	O
data	O
other	O
than	O
the	O
W3C	O
collection	O
.	O
At	O
the	O
same	O
time	O
,	O
many	O
intranets	O
of	O
universities	O
and	O
other	O
knowledge-intensive	O
organisations	O
offer	O
examples	O
of	O
relatively	O
small	O
but	O
clean	O
multilingual	O
expertise	O
data	O
,	O
covering	O
broad	O
ranges	O
of	O
expertise	O
areas	O
.	O
We	O
first	O
present	O
two	O
main	O
expertise	O
retrieval	O
tasks	O
,	O
along	O
with	O
a	O
set	O
of	O
baseline	O
approaches	O
based	O
on	O
generative	B-KEY
language	I-KEY
modeling	I-KEY
,	O
aimed	O
at	O
finding	O
expertise	O
relations	O
between	O
topics	O
and	O
people	O
.	O
For	O
our	O
experimental	O
evaluation	O
,	O
we	O
introduce	O
-LRB-	O
and	O
release	O
-RRB-	O
a	O
new	O
test	O
set	O
based	O
on	O
a	O
crawl	O
of	O
a	O
university	O
site	O
.	O
Using	O
this	O
test	O
set	O
,	O
we	O
conduct	O
two	O
series	O
of	O
experiments	O
.	O
The	O
first	O
is	O
aimed	O
at	O
determining	O
the	O
effectiveness	O
of	O
baseline	B-KEY
expertise	I-KEY
retrieval	I-KEY
methods	I-KEY
applied	O
to	O
the	O
new	O
test	O
set	O
.	O
The	O
second	O
is	O
aimed	O
at	O
assessing	O
refined	O
models	O
that	O
exploit	O
characteristic	O
features	O
of	O
the	O
new	O
test	O
set	O
,	O
such	O
as	O
the	O
organizational	B-KEY
structure	I-KEY
of	O
the	O
university	O
,	O
and	O
the	O
hierarchical	O
structure	O
of	O
the	O
topics	O
in	O
the	O
test	O
set	O
.	O
Expertise	O
retrieval	O
models	O
are	O
shown	O
to	O
be	O
robust	O
with	O
respect	O
to	O
environments	O
smaller	O
than	O
the	O
W3C	O
collection	O
,	O
and	O
current	O
techniques	O
appear	O
to	O
be	O
generalizable	O
to	O
other	O
settings	O
.	O

Evaluating	O
Adaptive	B-KEY
Resource	I-KEY
Management	I-KEY
for	O
Distributed	O
Real-Time	O
Embedded	O
Systems	O
ABSTRACT	O
A	O
challenging	O
problem	O
faced	O
by	O
researchers	O
and	O
developers	O
of	O
distributed	O
real-time	O
and	O
embedded	O
-LRB-	O
DRE	O
-RRB-	O
systems	O
is	O
devising	O
and	O
implementing	O
effective	O
adaptive	B-KEY
resource	I-KEY
management	I-KEY
strategies	O
that	O
can	O
meet	O
end-to-end	O
quality	B-KEY
of	I-KEY
service	I-KEY
-LRB-	O
QoS	O
-RRB-	O
requirements	O
in	O
varying	O
operational	O
conditions	O
.	O
This	O
paper	O
presents	O
two	O
contributions	O
to	O
research	O
in	O
adaptive	B-KEY
resource	I-KEY
management	I-KEY
for	O
DRE	O
systems	O
.	O
First	O
,	O
we	O
describe	O
the	O
structure	O
and	O
functionality	O
of	O
the	O
Hybrid	B-KEY
Adaptive	I-KEY
Resourcemanagement	I-KEY
Middleware	I-KEY
-LRB-	O
HyARM	O
-RRB-	O
,	O
which	O
provides	O
adaptive	B-KEY
resource	I-KEY
management	I-KEY
using	O
hybrid	B-KEY
control	I-KEY
techniques	I-KEY
for	O
adapting	O
to	O
workload	O
fluctuations	O
and	O
resource	O
availability	O
.	O
Second	O
,	O
we	O
evaluate	O
the	O
adaptive	O
behavior	O
of	O
HyARM	O
via	O
experiments	O
on	O
a	O
DRE	O
multimedia	O
system	O
that	O
distributes	O
video	O
in	O
real-time	O
.	O
Our	O
results	O
indicate	O
that	O
HyARM	O
yields	O
predictable	O
,	O
stable	O
,	O
and	O
high	O
system	O
performance	O
,	O
even	O
in	O
the	O
face	O
of	O
fluctuating	O
workload	O
and	O
resource	O
availability	O
.	O

Towards	O
Truthful	B-KEY
Mechanisms	I-KEY
for	O
Binary	B-KEY
Demand	I-KEY
Games	I-KEY
:	O
A	O
General	O
Framework	O
ABSTRACT	O
The	O
family	O
of	O
Vickrey-Clarke-Groves	O
-LRB-	O
VCG	O
-RRB-	O
mechanisms	O
is	O
arguably	O
the	O
most	O
celebrated	O
achievement	O
in	O
truthful	B-KEY
mechanism	I-KEY
design	O
.	O
However	O
,	O
VCG	O
mechanisms	O
have	O
their	O
limitations	O
.	O
They	O
only	O
apply	O
to	O
optimization	O
problems	O
with	O
a	O
utilitarian	O
-LRB-	O
or	O
affine	O
-RRB-	O
objective	B-KEY
function	I-KEY
,	O
and	O
their	O
output	O
should	O
optimize	O
the	O
objective	B-KEY
function	I-KEY
.	O
For	O
many	O
optimization	O
problems	O
,	O
finding	O
the	O
optimal	O
output	O
is	O
computationally	O
intractable	O
.	O
If	O
we	O
apply	O
VCG	O
mechanisms	O
to	O
polynomial-time	O
algorithms	O
that	O
approximate	O
the	O
optimal	O
solution	O
,	O
the	O
resulting	O
mechanisms	O
may	O
no	O
longer	O
be	O
truthful	O
.	O
In	O
light	O
of	O
these	O
limitations	O
,	O
it	O
is	O
useful	O
to	O
study	O
whether	O
we	O
can	O
design	O
a	O
truthful	O
non-VCG	O
payment	O
scheme	O
that	O
is	O
computationally	O
tractable	O
for	O
a	O
given	O
allocation	O
rule	O
O	O
.	O
In	O
this	O
paper	O
,	O
we	O
focus	O
our	O
attention	O
on	O
binary	B-KEY
demand	I-KEY
games	I-KEY
in	O
which	O
the	O
agents	O
'	O
only	O
available	O
actions	O
are	O
to	O
take	O
part	O
in	O
the	O
a	O
game	O
or	O
not	O
to	O
.	O
For	O
these	O
problems	O
,	O
we	O
prove	O
that	O
a	O
truthful	B-KEY
mechanism	I-KEY
M	O
=	O
-LRB-	O
O	O
,	O
P	O
-RRB-	O
exists	O
with	O
a	O
proper	O
payment	O
method	O
P	O
iff	O
the	O
allocation	O
rule	O
O	O
satisfies	O
a	O
certain	O
monotonicity	B-KEY
property	I-KEY
.	O
We	O
provide	O
a	O
general	O
framework	O
to	O
design	O
such	O
P	O
.	O
We	O
further	O
propose	O
several	O
general	O
composition-based	O
techniques	O
to	O
compute	O
P	O
efficiently	O
for	O
various	O
types	O
of	O
output	O
.	O
In	O
particular	O
,	O
we	O
show	O
how	O
P	O
can	O
be	O
computed	O
through	O
``	O
or/and	O
''	O
combinations	B-KEY
,	O
round-based	O
combinations	B-KEY
,	O
and	O
some	O
more	O
complex	O
combinations	B-KEY
of	O
the	O
outputs	O
from	O
subgames	O
.	O

Exchanging	O
Reputation	B-KEY
Values	O
among	O
Heterogeneous	O
Agent	O
Reputation	O
Models	O
:	O
An	O
Experience	O
on	O
ART	O
Testbed	O
ABSTRACT	O
In	O
open	O
MAS	O
it	O
is	O
often	O
a	O
problem	O
to	O
achieve	O
agents	O
'	O
interoperability	B-KEY
.	O
The	O
heterogeneity	O
of	O
its	O
components	O
turns	O
the	O
establishment	O
of	O
interaction	O
or	O
cooperation	O
among	O
them	O
into	O
a	O
non	O
trivial	O
task	O
,	O
since	O
agents	O
may	O
use	O
different	O
internal	O
models	O
and	O
the	O
decision	O
about	O
trust	B-KEY
other	O
agents	O
is	O
a	O
crucial	O
condition	O
to	O
the	O
formation	O
of	O
agents	O
'	O
cooperation	O
.	O
In	O
this	O
paper	O
we	O
propose	O
the	O
use	O
of	O
an	O
ontology	B-KEY
to	O
deal	O
with	O
this	O
issue	O
.	O
We	O
experiment	O
this	O
idea	O
by	O
enhancing	O
the	O
ART	O
reputation	B-KEY
model	O
with	O
semantic	O
data	O
obtained	O
from	O
this	O
ontology	O
.	O
This	O
data	O
is	O
used	O
during	O
interaction	O
among	O
heterogeneous	B-KEY
agents	I-KEY
when	O
exchanging	O
reputation	B-KEY
values	O
and	O
may	O
be	O
used	O
for	O
agents	O
that	O
use	O
different	O
reputation	O
models	O
.	O

Negotiation	B-KEY
by	O
Abduction	O
and	O
Relaxation	B-KEY
ABSTRACT	O
This	O
paper	O
studies	O
a	O
logical	O
framework	O
for	O
automated	B-KEY
negotiation	I-KEY
between	O
two	O
agents	O
.	O
We	O
suppose	O
an	O
agent	O
who	O
has	O
a	O
knowledge	O
base	O
represented	O
by	O
a	O
logic	B-KEY
program	I-KEY
.	O
Then	O
,	O
we	O
introduce	O
methods	O
of	O
constructing	O
counter-proposals	O
in	O
response	O
to	O
proposals	O
made	O
by	O
an	O
agent	O
.	O
To	O
this	O
end	O
,	O
we	O
combine	O
the	O
techniques	O
of	O
extended	B-KEY
abduction	I-KEY
in	O
artificial	O
intelligence	O
and	O
relaxation	B-KEY
in	O
cooperative	O
query	O
answering	O
for	O
databases	O
.	O
These	O
techniques	O
are	O
respectively	O
used	O
for	O
producing	O
conditional	B-KEY
proposals	I-KEY
and	O
neighborhood	O
proposals	O
in	O
the	O
process	O
of	O
negotiation	B-KEY
.	O
We	O
provide	O
a	O
negotiation	B-KEY
protocol	O
based	O
on	O
the	O
exchange	O
of	O
these	O
proposals	O
and	O
develop	O
procedures	O
for	O
computing	O
new	O
proposals	O
.	O

On	O
the	O
Computational	O
Power	O
of	O
Iterative	O
Auctions	O
*	O
ABSTRACT	O
We	O
embark	O
on	O
a	O
systematic	O
analysis	O
of	O
the	O
power	O
and	O
limitations	O
of	O
iterative	O
combinatorial	B-KEY
auctions	I-KEY
.	O
Most	O
existing	O
iterative	O
combinatorial	B-KEY
auctions	I-KEY
are	O
based	O
on	O
repeatedly	O
suggesting	O
prices	B-KEY
for	O
bundles	O
of	O
items	O
,	O
and	O
querying	O
the	O
bidders	B-KEY
for	O
their	O
``	O
demand	O
''	O
under	O
these	O
prices	B-KEY
.	O
We	O
prove	O
a	O
large	O
number	O
of	O
results	O
showing	O
the	O
boundaries	O
of	O
what	O
can	O
be	O
achieved	O
by	O
auctions	O
of	O
this	O
kind	O
.	O
We	O
first	O
focus	O
on	O
auctions	O
that	O
use	O
a	O
polynomial	O
number	O
of	O
demand	B-KEY
queries	I-KEY
,	O
and	O
then	O
we	O
analyze	O
the	O
power	O
of	O
different	O
kinds	O
of	O
ascending-price	O
auctions	O
.	O

Information	B-KEY
Markets	I-KEY
vs.	O
Opinion	B-KEY
Pools	I-KEY
:	O
An	O
Empirical	O
Comparison	O
ABSTRACT	O
In	O
this	O
paper	O
,	O
we	O
examine	O
the	O
relative	O
forecast	B-KEY
accuracy	O
of	O
information	B-KEY
markets	I-KEY
versus	O
expert	B-KEY
aggregation	I-KEY
.	O
We	O
leverage	O
a	O
unique	O
data	O
source	O
of	O
almost	O
2000	O
people	O
's	O
subjective	O
probability	O
judgments	O
on	O
2003	O
US	O
National	O
Football	O
League	O
games	O
and	O
compare	O
with	O
the	O
``	O
market	B-KEY
probabilities	I-KEY
''	O
given	O
by	O
two	O
different	O
information	B-KEY
markets	I-KEY
on	O
exactly	O
the	O
same	O
events	O
.	O
We	O
combine	O
assessments	O
of	O
multiple	O
experts	O
via	O
linear	O
and	O
logarithmic	O
aggregation	O
functions	O
to	O
form	O
pooled	B-KEY
predictions	I-KEY
.	O
Prices	B-KEY
in	O
information	B-KEY
markets	I-KEY
are	O
used	O
to	O
derive	O
market	O
predictions	O
.	O
Our	O
results	O
show	O
that	O
,	O
at	O
the	O
same	O
time	O
point	O
ahead	O
of	O
the	O
game	O
,	O
information	B-KEY
markets	I-KEY
provide	O
as	O
accurate	O
predictions	O
as	O
pooled	O
expert	O
assessments	O
.	O
In	O
screening	O
pooled	O
expert	O
predictions	O
,	O
we	O
find	O
that	O
arithmetic	O
average	O
is	O
a	O
robust	O
and	O
efficient	O
pooling	O
function	O
;	O
weighting	O
expert	O
assessments	O
according	O
to	O
their	O
past	O
performance	O
does	O
not	O
improve	O
accuracy	O
of	O
pooled	B-KEY
predictions	I-KEY
;	O
and	O
logarithmic	O
aggregation	O
functions	O
offer	O
bolder	O
predictions	O
than	O
linear	O
aggregation	O
functions	O
.	O
The	O
results	O
provide	O
insights	O
into	O
the	O
predictive	O
performance	O
of	O
information	B-KEY
markets	I-KEY
,	O
and	O
the	O
relative	O
merits	O
of	O
selecting	O
among	O
various	O
opinion	B-KEY
pooling	I-KEY
methods	O
.	O

A	O
Price-Anticipating	O
Resource	B-KEY
Allocation	I-KEY
Mechanism	O
for	O
Distributed	B-KEY
Shared	I-KEY
Clusters	I-KEY
ABSTRACT	O
In	O
this	O
paper	O
we	O
formulate	O
the	O
fixed	O
budget	O
resource	B-KEY
allocation	I-KEY
game	O
to	O
understand	O
the	O
performance	O
of	O
a	O
distributed	O
marketbased	O
resource	B-KEY
allocation	I-KEY
system	O
.	O
Multiple	O
users	O
decide	O
how	O
to	O
distribute	O
their	O
budget	O
-LRB-	O
bids	O
-RRB-	O
among	O
multiple	O
machines	O
according	O
to	O
their	O
individual	O
preferences	O
to	O
maximize	O
their	O
individual	O
utility	B-KEY
.	O
We	O
look	O
at	O
both	O
the	O
efficiency	B-KEY
and	O
the	O
fairness	B-KEY
of	O
the	O
allocation	O
at	O
the	O
equilibrium	O
,	O
where	O
fairness	B-KEY
is	O
evaluated	O
through	O
the	O
measures	O
of	O
utility	B-KEY
uniformity	O
and	O
envy-freeness	O
.	O
We	O
show	O
analytically	O
and	O
through	O
simulations	B-KEY
that	O
despite	O
being	O
highly	O
decentralized	O
,	O
such	O
a	O
system	O
converges	O
quickly	O
to	O
an	O
equilibrium	O
and	O
unlike	O
the	O
social	O
optimum	O
that	O
achieves	O
high	O
efficiency	B-KEY
but	O
poor	O
fairness	B-KEY
,	O
the	O
proposed	O
allocation	O
scheme	O
achieves	O
a	O
nice	O
balance	O
of	O
high	O
degrees	O
of	O
efficiency	B-KEY
and	O
fairness	B-KEY
at	O
the	O
equilibrium	O
.	O

TSAR	O
:	O
A	O
Two	O
Tier	O
Sensor	O
Storage	O
Architecture	O
Using	O
Interval	B-KEY
Skip	I-KEY
Graphs	I-KEY
*	O
ABSTRACT	O
Archival	B-KEY
storage	O
of	O
sensor	O
data	O
is	O
necessary	O
for	O
applications	O
that	O
query	O
,	O
mine	O
,	O
and	O
analyze	O
such	O
data	O
for	O
interesting	O
features	O
and	O
trends	O
.	O
We	O
argue	O
that	O
existing	O
storage	O
systems	O
are	O
designed	O
primarily	O
for	O
flat	O
hierarchies	O
of	O
homogeneous	O
sensor	O
nodes	O
and	O
do	O
not	O
fully	O
exploit	O
the	O
multi-tier	O
nature	O
of	O
emerging	O
sensor	O
networks	O
,	O
where	O
an	O
application	O
can	O
comprise	O
tens	O
of	O
tethered	O
proxies	O
,	O
each	O
managing	O
tens	O
to	O
hundreds	O
of	O
untethered	O
sensors	O
.	O
We	O
present	O
TSAR	O
,	O
a	O
fundamentally	O
different	O
storage	O
architecture	O
that	O
envisions	O
separation	B-KEY
of	I-KEY
data	I-KEY
from	O
metadata	O
by	O
employing	O
local	O
archiving	B-KEY
at	O
the	O
sensors	O
and	O
distributed	O
indexing	O
at	O
the	O
proxies	O
.	O
At	O
the	O
proxy	O
tier	O
,	O
TSAR	O
employs	O
a	O
novel	O
multi-resolution	O
ordered	O
distributed	B-KEY
index	I-KEY
structure	I-KEY
,	O
the	O
Interval	B-KEY
Skip	I-KEY
Graph	I-KEY
,	O
for	O
efficiently	O
supporting	O
spatio-temporal	O
and	O
value	O
queries	O
.	O
At	O
the	O
sensor	O
tier	O
,	O
TSAR	O
supports	O
energy-aware	O
adaptive	O
summarization	O
that	O
can	O
trade	O
off	O
the	O
cost	O
of	O
transmitting	O
metadata	O
to	O
the	O
proxies	O
against	O
the	O
overhead	O
offalse	O
hits	O
resultingfrom	O
querying	O
a	O
coarse-grain	O
index	O
.	O
We	O
implement	O
TSAR	O
in	O
a	O
two-tier	O
sensor	O
testbed	O
comprising	O
Stargatebased	O
proxies	O
and	O
Mote-based	O
sensors	O
.	O
Our	O
experiments	O
demonstrate	O
the	O
benefits	O
and	O
feasibility	O
of	O
using	O
our	O
energy-efficient	O
storage	O
architecture	O
in	O
multi-tier	O
sensor	O
networks	O
.	O

Marginal	B-KEY
Contribution	I-KEY
Nets	O
:	O
A	O
Compact	O
Representation	B-KEY
Scheme	O
for	O
Coalitional	O
Games	O
*	O
ABSTRACT	O
We	O
present	O
a	O
new	O
approach	O
to	O
representing	O
coalitional	B-KEY
games	I-KEY
based	O
on	O
rules	O
that	O
describe	O
the	O
marginal	B-KEY
contributions	I-KEY
of	O
the	O
agents	B-KEY
.	O
This	O
representation	B-KEY
scheme	O
captures	O
characteristics	O
of	O
the	O
interactions	B-KEY
among	O
the	O
agents	B-KEY
in	O
a	O
natural	O
and	O
concise	O
manner	O
.	O
We	O
also	O
develop	O
efficient	O
algorithms	O
for	O
two	O
of	O
the	O
most	O
important	O
solution	O
concepts	O
,	O
the	O
Shapley	O
value	O
and	O
the	O
core	B-KEY
,	O
under	O
this	O
representation	B-KEY
.	O
The	O
Shapley	O
value	O
can	O
be	O
computed	O
in	O
time	O
linear	O
in	O
the	O
size	O
of	O
the	O
input	O
.	O
The	O
emptiness	O
of	O
the	O
core	B-KEY
can	O
be	O
determined	O
in	O
time	O
exponential	O
only	O
in	O
the	O
treewidth	B-KEY
of	O
a	O
graphical	O
interpretation	O
of	O
our	O
representation	B-KEY
.	O

Demonstration	O
of	O
Grid-Enabled	O
Ensemble	B-KEY
Kalman	I-KEY
Filter	I-KEY
Data	B-KEY
Assimilation	I-KEY
Methodology	I-KEY
for	O
Reservoir	O
Characterization	O
ABSTRACT	O
Ensemble	B-KEY
Kalman	I-KEY
filter	I-KEY
data	B-KEY
assimilation	I-KEY
methodology	I-KEY
is	O
a	O
popular	O
approach	O
for	O
hydrocarbon	B-KEY
reservoir	I-KEY
simulations	I-KEY
in	O
energy	B-KEY
exploration	I-KEY
.	O
In	O
this	O
approach	O
,	O
an	O
ensemble	O
of	O
geological	O
models	O
and	O
production	O
data	O
of	O
oil	O
fields	O
is	O
used	O
to	O
forecast	O
the	O
dynamic	O
response	O
of	O
oil	O
wells	O
.	O
The	O
Schlumberger	O
ECLIPSE	O
software	O
is	O
used	O
for	O
these	O
simulations	O
.	O
Since	O
models	O
in	O
the	O
ensemble	O
do	O
not	O
communicate	O
,	O
message-passing	O
implementation	O
is	O
a	O
good	O
choice	O
.	O
Each	O
model	O
checks	O
out	O
an	O
ECLIPSE	O
license	O
and	O
therefore	O
,	O
parallelizability	O
of	O
reservoir	O
simulations	O
depends	O
on	O
the	O
number	O
licenses	O
available	O
.	O
We	O
have	O
Grid-enabled	O
the	O
ensemble	B-KEY
Kalman	I-KEY
filter	I-KEY
data	B-KEY
assimilation	I-KEY
methodology	I-KEY
for	O
the	O
TIGRE	B-KEY
Grid	B-KEY
computing	I-KEY
environment	O
.	O
By	O
pooling	O
the	O
licenses	O
and	O
computing	O
resources	O
across	O
the	O
collaborating	O
institutions	O
using	O
GridWay	O
metascheduler	O
and	O
TIGRE	B-KEY
environment	O
,	O
the	O
computational	O
accuracy	O
can	O
be	O
increased	O
while	O
reducing	O
the	O
simulation	O
runtime	O
.	O
In	O
this	O
paper	O
,	O
we	O
provide	O
an	O
account	O
of	O
our	O
efforts	O
in	O
Gridenabling	O
the	O
ensemble	B-KEY
Kalman	I-KEY
Filter	I-KEY
data	B-KEY
assimilation	I-KEY
methodology	I-KEY
.	O
Potential	O
benefits	O
of	O
this	O
approach	O
,	O
observations	O
and	O
lessons	O
learned	O
will	O
be	O
discussed	O
.	O

A	O
Unified	O
and	O
General	O
Framework	B-KEY
for	O
Argumentation-based	O
Negotiation	B-KEY
ABSTRACT	O
This	O
paper	O
proposes	O
a	O
unified	O
and	O
general	O
framework	B-KEY
for	O
argumentation-based	O
negotiation	B-KEY
,	O
in	O
which	O
the	O
role	O
of	O
argumentation	B-KEY
is	O
formally	O
analyzed	O
.	O
The	O
framework	B-KEY
makes	O
it	O
possible	O
to	O
study	O
the	O
outcomes	B-KEY
of	O
an	O
argumentation-based	O
negotiation	B-KEY
.	O
It	O
shows	O
what	O
an	O
agreement	O
is	O
,	O
how	O
it	O
is	O
related	O
to	O
the	O
theories	B-KEY
of	O
the	O
agents	B-KEY
,	O
when	O
it	O
is	O
possible	O
,	O
and	O
how	O
this	O
can	O
be	O
attained	O
by	O
the	O
negotiating	B-KEY
agents	B-KEY
in	O
this	O
case	O
.	O
It	O
defines	O
also	O
the	O
notion	B-KEY
of	I-KEY
concession	I-KEY
,	O
and	O
shows	O
in	O
which	O
situation	O
an	O
agent	B-KEY
will	O
make	O
one	O
,	O
as	O
well	O
as	O
how	O
it	O
influences	O
the	O
evolution	O
of	O
the	O
dialogue	O
.	O

The	O
Dynamics	O
of	O
Viral	B-KEY
Marketing	I-KEY
*	O
ABSTRACT	O
We	O
present	O
an	O
analysis	O
of	O
a	O
person-to-person	O
recommendation	B-KEY
network	I-KEY
,	O
consisting	O
of	O
4	O
million	O
people	O
who	O
made	O
16	O
million	O
recommendations	O
on	O
half	O
a	O
million	O
products	B-KEY
.	O
We	O
observe	O
the	O
propagation	O
of	O
recommendations	O
and	O
the	O
cascade	O
sizes	O
,	O
which	O
we	O
explain	O
by	O
a	O
simple	O
stochastic	B-KEY
model	I-KEY
.	O
We	O
then	O
establish	O
how	O
the	O
recommendation	B-KEY
network	I-KEY
grows	O
over	O
time	O
and	O
how	O
effective	O
it	O
is	O
from	O
the	O
viewpoint	O
of	O
the	O
sender	O
and	O
receiver	O
of	O
the	O
recommendations	O
.	O
While	O
on	O
average	O
recommendations	O
are	O
not	O
very	O
effective	O
at	O
inducing	O
purchases	B-KEY
and	O
do	O
not	O
spread	O
very	O
far	O
,	O
we	O
present	O
a	O
model	O
that	O
successfully	O
identifies	O
product	B-KEY
and	O
pricing	B-KEY
categories	I-KEY
for	O
which	O
viral	B-KEY
marketing	I-KEY
seems	O
to	O
be	O
very	O
effective	O
.	O

Implementing	O
Commitment-Based	O
Interactions	O
*	O
ABSTRACT	O
Although	O
agent	B-KEY
interaction	I-KEY
plays	O
a	O
vital	O
role	O
in	O
MAS	O
,	O
and	O
messagecentric	B-KEY
approaches	I-KEY
to	O
agent	B-KEY
interaction	I-KEY
have	O
their	O
drawbacks	O
,	O
present	O
agent-oriented	O
programming	O
languages	O
do	O
not	O
provide	O
support	O
for	O
implementing	O
agent	B-KEY
interaction	I-KEY
that	O
is	O
flexible	O
and	O
robust	O
.	O
Instead	O
,	O
messages	O
are	O
provided	O
as	O
a	O
primitive	O
building	O
block	O
.	O
In	O
this	O
paper	O
we	O
consider	O
one	O
approach	O
for	O
modelling	O
agent	B-KEY
interactions	I-KEY
:	O
the	O
commitment	B-KEY
machines	I-KEY
framework	O
.	O
This	O
framework	O
supports	O
modelling	O
interactions	O
at	O
a	O
higher	O
level	O
-LRB-	O
using	O
social	B-KEY
commitments	I-KEY
-RRB-	O
,	O
resulting	O
in	O
more	O
flexible	O
interactions	O
.	O
We	O
investigate	O
how	O
commitmentbased	O
interactions	O
can	O
be	O
implemented	O
in	O
conventional	O
agent-oriented	O
programming	O
languages	O
.	O
The	O
contributions	O
of	O
this	O
paper	O
are	O
:	O
a	O
mapping	O
from	O
a	O
commitment	B-KEY
machine	I-KEY
to	O
a	O
collection	O
of	O
BDI-style	O
plans	O
;	O
extensions	O
to	O
the	O
semantics	O
of	O
BDI	O
programming	O
languages	O
;	O
and	O
an	O
examination	O
of	O
two	O
issues	O
that	O
arise	O
when	O
distributing	O
commitment	O
machines	O
-LRB-	O
turn	O
management	O
and	O
race	O
conditions	O
-RRB-	O
and	O
solutions	O
to	O
these	O
problems	O
.	O

On	O
Decentralized	B-KEY
Incentive	I-KEY
Compatible	I-KEY
Mechanisms	I-KEY
for	O
Partially	B-KEY
Informed	I-KEY
Environments	I-KEY
*	O
ABSTRACT	O
Algorithmic	O
Mechanism	O
Design	O
focuses	O
on	O
Dominant	B-KEY
Strategy	I-KEY
Implementations	I-KEY
.	O
The	O
main	O
positive	O
results	O
are	O
the	O
celebrated	O
Vickrey-Clarke-Groves	O
-LRB-	O
VCG	O
-RRB-	O
mechanisms	O
and	O
computationally	O
efficient	O
mechanisms	O
for	O
severely	O
restricted	O
players	O
-LRB-	O
``	O
single-parameter	O
domains	O
''	O
-RRB-	O
.	O
As	O
it	O
turns	O
out	O
,	O
many	O
natural	O
social	O
goals	O
can	O
not	O
be	O
implemented	O
using	O
the	O
dominant	O
strategy	O
concept	O
-LSB-	O
35	O
,	O
32	O
,	O
22	O
,	O
20	O
-RSB-	O
.	O
This	O
suggests	O
that	O
the	O
standard	O
requirements	O
must	O
be	O
relaxed	O
in	O
order	O
to	O
construct	O
general-purpose	O
mechanisms	O
.	O
We	O
observe	O
that	O
in	O
many	O
common	O
distributed	B-KEY
environments	I-KEY
computational	B-KEY
entities	I-KEY
can	O
take	O
advantage	O
of	O
the	O
network	O
structure	O
to	O
collect	O
and	O
distribute	O
information	O
.	O
We	O
thus	O
suggest	O
a	O
notion	O
of	O
partially	B-KEY
informed	I-KEY
environments	I-KEY
.	O
Even	O
if	O
the	O
information	O
is	O
recorded	O
with	O
some	O
probability	O
,	O
this	O
enables	O
us	O
to	O
implement	O
a	O
wider	O
range	O
of	O
social	O
goals	O
,	O
using	O
the	O
concept	O
of	O
iterative	B-KEY
elimination	I-KEY
of	I-KEY
weakly	I-KEY
dominated	I-KEY
strategies	I-KEY
.	O
As	O
a	O
result	O
,	O
cooperation	B-KEY
is	O
achieved	O
independent	O
of	O
agents	B-KEY
'	O
belief	O
.	O
As	O
a	O
case	O
study	O
,	O
we	O
apply	O
our	O
methods	O
to	O
derive	O
Peer-to-Peer	O
network	O
mechanism	O
for	O
file	O
sharing	O
.	O

Consistency-preserving	O
Caching	O
of	O
Dynamic	O
Database	B-KEY
Content	I-KEY
*	O
ABSTRACT	O
With	O
the	O
growing	O
use	O
of	O
dynamic	O
web	O
content	O
generated	O
from	O
relational	B-KEY
databases	I-KEY
,	O
traditional	O
caching	O
solutions	O
for	O
throughput	O
and	O
latency	O
improvements	O
are	O
ineffective	O
.	O
We	O
describe	O
a	O
middleware	O
layer	O
called	O
Ganesh	O
that	O
reduces	O
the	O
volume	O
of	O
data	O
transmitted	O
without	O
semantic	O
interpretation	O
of	O
queries	O
or	O
results	O
.	O
It	O
achieves	O
this	O
reduction	O
through	O
the	O
use	O
of	O
cryptographic	O
hashing	O
to	O
detect	O
similarities	O
with	O
previous	O
results	O
.	O
These	O
benefits	O
do	O
not	O
require	O
any	O
compromise	O
of	O
the	O
strict	O
consistency	O
semantics	O
provided	O
by	O
the	O
back-end	O
database	O
.	O
Further	O
,	O
Ganesh	O
does	O
not	O
require	O
modifications	O
to	O
applications	O
,	O
web	O
servers	O
,	O
or	O
database	O
servers	O
,	O
and	O
works	O
with	O
closed-source	O
applications	O
and	O
databases	O
.	O
Using	O
two	O
benchmarks	O
representative	O
of	O
dynamic	O
web	O
sites	O
,	O
measurements	O
of	O
our	O
prototype	O
show	O
that	O
it	O
can	O
increase	O
end-to-end	O
throughput	O
by	O
as	O
much	O
as	O
twofold	O
for	O
non-data	O
intensive	O
applications	O
and	O
by	O
as	O
much	O
as	O
tenfold	O
for	O
data	O
intensive	O
ones	O
.	O

Adapting	O
Asynchronous	B-KEY
Messaging	I-KEY
Middleware	I-KEY
to	O
Ad	O
Hoc	O
Networking	O
ABSTRACT	O
The	O
characteristics	O
of	O
mobile	O
environments	O
,	O
with	O
the	O
possibility	O
of	O
frequent	O
disconnections	O
and	O
fluctuating	O
bandwidth	O
,	O
have	O
forced	O
a	O
rethink	O
of	O
traditional	O
middleware	O
.	O
In	O
particular	O
,	O
the	O
synchronous	O
communication	O
paradigms	O
often	O
employed	O
in	O
standard	O
middleware	O
do	O
not	O
appear	O
to	O
be	O
particularly	O
suited	O
to	O
ad	O
hoc	O
environments	O
,	O
in	O
which	O
not	O
even	O
the	O
intermittent	O
availability	O
of	O
a	O
backbone	O
network	O
can	O
be	O
assumed	O
.	O
Instead	O
,	O
asynchronous	B-KEY
communication	I-KEY
seems	O
to	O
be	O
a	O
generally	O
more	O
suitable	O
paradigm	O
for	O
such	O
environments	O
.	O
Message	B-KEY
oriented	I-KEY
middleware	I-KEY
for	O
traditional	O
systems	O
has	O
been	O
developed	O
and	O
used	O
to	O
provide	O
an	O
asynchronous	O
paradigm	O
of	O
communication	O
for	O
distributed	O
systems	O
,	O
and	O
,	O
recently	O
,	O
also	O
for	O
some	O
specific	O
mobile	O
computing	O
systems	O
.	O
In	O
this	O
paper	O
,	O
we	O
present	O
our	O
experience	O
in	O
designing	O
,	O
implementing	O
and	O
evaluating	O
EMMA	O
-LRB-	O
Epidemic	B-KEY
Messaging	I-KEY
Middleware	I-KEY
for	O
Ad	O
hoc	O
networks	O
-RRB-	O
,	O
an	O
adaptation	O
of	O
Java	B-KEY
Message	I-KEY
Service	I-KEY
-LRB-	O
JMS	O
-RRB-	O
for	O
mobile	O
ad	O
hoc	O
environments	O
.	O
We	O
discuss	O
in	O
detail	O
the	O
design	O
challenges	O
and	O
some	O
possible	O
solutions	O
,	O
showing	O
a	O
concrete	O
example	O
of	O
the	O
feasibility	O
and	O
suitability	O
of	O
the	O
application	O
of	O
the	O
asynchronous	O
paradigm	O
in	O
this	O
setting	O
and	O
outlining	O
a	O
research	O
roadmap	O
for	O
the	O
coming	O
years	O
.	O

A	O
Hierarchical	B-KEY
Process	I-KEY
Execution	I-KEY
Support	O
for	O
Grid	O
Computing	O
ABSTRACT	O
Grid	O
is	O
an	O
emerging	O
infrastructure	O
used	O
to	O
share	O
resources	O
among	O
virtual	O
organizations	O
in	O
a	O
seamless	O
manner	O
and	O
to	O
provide	O
breakthrough	O
computing	O
power	O
at	O
low	O
cost	O
.	O
Nowadays	O
there	O
are	O
dozens	O
of	O
academic	O
and	O
commercial	O
products	O
that	O
allow	O
execution	O
of	O
isolated	O
tasks	O
on	O
grids	O
,	O
but	O
few	O
products	O
support	O
the	O
enactment	O
of	O
long-running	O
processes	O
in	O
a	O
distributed	O
fashion	O
.	O
In	O
order	O
to	O
address	O
such	O
subject	O
,	O
this	O
paper	O
presents	O
a	O
programming	O
model	O
and	O
an	O
infrastructure	O
that	O
hierarchically	O
schedules	O
process	O
activities	O
using	O
available	O
nodes	O
in	O
a	O
wide	O
grid	O
environment	O
.	O
Their	O
advantages	O
are	O
automatic	O
and	O
structured	O
distribution	O
of	O
activities	O
and	O
easy	O
process	O
monitoring	O
and	O
steering	O
.	O

Machine	B-KEY
Learning	I-KEY
for	O
Information	B-KEY
Architecture	I-KEY
in	O
a	O
Large	O
Governmental	O
Website	O
*	O
ABSTRACT	O
This	O
paper	O
describes	O
ongoing	O
research	O
into	O
the	O
application	O
of	O
machine	B-KEY
learning	I-KEY
techniques	O
for	O
improving	O
access	O
to	O
governmental	O
information	O
in	O
complex	O
digital	O
libraries	O
.	O
Under	O
the	O
auspices	O
of	O
the	O
GovStat	O
Project	O
,	O
our	O
goal	O
is	O
to	O
identify	O
a	O
small	O
number	O
of	O
semantically	O
valid	O
concepts	O
that	O
adequately	O
spans	O
the	O
intellectual	O
domain	O
of	O
a	O
collection	O
.	O
The	O
goal	O
of	O
this	O
discovery	O
is	O
twofold	O
.	O
First	O
we	O
desire	O
a	O
practical	O
aid	O
for	O
information	O
architects	O
.	O
Second	O
,	O
automatically	O
derived	O
documentconcept	O
relationships	O
are	O
a	O
necessary	O
precondition	O
for	O
realworld	O
deployment	O
of	O
many	O
dynamic	O
interfaces	O
.	O
The	O
current	O
study	O
compares	O
concept	O
learning	O
strategies	O
based	O
on	O
three	O
document	O
representations	O
:	O
keywords	O
,	O
titles	O
,	O
and	O
full-text	O
.	O
In	O
statistical	O
and	O
user-based	O
studies	O
,	O
human-created	O
keywords	O
provide	O
significant	O
improvements	O
in	O
concept	O
learning	O
over	O
both	O
title-only	O
and	O
full-text	O
representations	O
.	O

Network	B-KEY
Monitors	I-KEY
and	O
Contracting	O
Systems	O
:	O
Competition	O
and	O
Innovation	O
ABSTRACT	O
Today	O
's	O
Internet	O
industry	O
suffers	O
from	O
several	O
well-known	O
pathologies	O
,	O
but	O
none	O
is	O
as	O
destructive	O
in	O
the	O
long	O
term	O
as	O
its	O
resistance	O
to	O
evolution	O
.	O
Rather	O
than	O
introducing	O
new	O
services	O
,	O
ISPs	O
are	O
presently	O
moving	O
towards	O
greater	O
commoditization	B-KEY
.	O
It	O
is	O
apparent	O
that	O
the	O
network	O
's	O
primitive	O
system	O
of	O
contracts	B-KEY
does	O
not	O
align	O
incentives	B-KEY
properly	O
.	O
In	O
this	O
study	O
,	O
we	O
identify	O
the	O
network	O
's	O
lack	O
of	O
accountability	O
as	O
a	O
fundamental	O
obstacle	O
to	O
correcting	O
this	O
problem	O
:	O
Employing	O
an	O
economic	O
model	O
,	O
we	O
argue	O
that	O
optimal	O
routes	O
and	O
innovation	B-KEY
are	O
impossible	O
unless	O
new	O
monitoring	B-KEY
capability	O
is	O
introduced	O
and	O
incorporated	O
with	O
the	O
contracting	B-KEY
system	O
.	O
Furthermore	O
,	O
we	O
derive	O
the	O
minimum	O
requirements	O
a	O
monitoring	B-KEY
system	O
must	O
meet	O
to	O
support	O
first-best	O
routing	O
and	O
innovation	B-KEY
characteristics	O
.	O
Our	O
work	O
does	O
not	O
constitute	O
a	O
new	O
protocol	O
;	O
rather	O
,	O
we	O
provide	O
practical	O
and	O
specific	O
guidance	O
for	O
the	O
design	O
of	O
monitoring	B-KEY
systems	O
,	O
as	O
well	O
as	O
a	O
theoretical	O
framework	O
to	O
explore	O
the	O
factors	O
that	O
influence	O
innovation	B-KEY
.	O

Agents	O
,	O
Beliefs	B-KEY
,	O
and	O
Plausible	B-KEY
Behavior	O
in	O
a	O
Temporal	O
Setting	O
ABSTRACT	O
Logics	B-KEY
of	O
knowledge	O
and	O
belief	B-KEY
are	O
often	O
too	O
static	O
and	O
inflexible	O
to	O
be	O
used	O
on	O
real-world	O
problems	O
.	O
In	O
particular	O
,	O
they	O
usually	O
offer	O
no	O
concept	O
for	O
expressing	O
that	O
some	O
course	O
of	O
events	O
is	O
more	O
likely	O
to	O
happen	O
than	O
another	O
.	O
We	O
address	O
this	O
problem	O
and	O
extend	O
CTLK	O
-LRB-	O
computation	B-KEY
tree	I-KEY
logic	I-KEY
with	O
knowledge	O
-RRB-	O
with	O
a	O
notion	O
of	O
plausibility	O
,	O
which	O
allows	O
for	O
practical	O
and	O
counterfactual	O
reasoning	O
.	O
The	O
new	O
logic	B-KEY
CTLKP	O
-LRB-	O
CTLK	O
with	O
plausibility	B-KEY
-RRB-	O
includes	O
also	O
a	O
particular	O
notion	B-KEY
of	I-KEY
belief	I-KEY
.	O
A	O
plausibility	B-KEY
update	O
operator	O
is	O
added	O
to	O
this	O
logic	O
in	O
order	O
to	O
change	O
plausibility	O
assumptions	O
dynamically	O
.	O
Furthermore	O
,	O
we	O
examine	O
some	O
important	O
properties	O
of	O
these	O
concepts	O
.	O
In	O
particular	O
,	O
we	O
show	O
that	O
,	O
for	O
a	O
natural	O
class	O
of	O
models	O
,	O
belief	B-KEY
is	O
a	O
KD45	O
modality	O
.	O
We	O
also	O
show	O
that	O
model	O
checking	O
CTLKP	O
is	O
PTIME-complete	O
and	O
can	O
be	O
done	O
in	O
time	O
linear	O
with	O
respect	O
to	O
the	O
size	O
of	O
models	O
and	O
formulae	O
.	O

Modular	B-KEY
Interpreted	I-KEY
Systems	I-KEY
ABSTRACT	O
We	O
propose	O
a	O
new	O
class	O
of	O
representations	O
that	O
can	O
be	O
used	O
for	O
modeling	O
-LRB-	O
and	O
model	B-KEY
checking	I-KEY
-RRB-	O
temporal	O
,	O
strategic	O
and	O
epistemic	O
properties	O
of	O
agents	O
and	O
their	O
teams	O
.	O
Our	O
representations	O
borrow	O
the	O
main	O
ideas	O
from	O
interpreted	O
systems	O
of	O
Halpern	O
,	O
Fagin	O
et	O
al.	O
;	O
however	O
,	O
they	O
are	O
also	O
modular	O
and	O
compact	O
in	O
the	O
way	O
concurrent	O
programs	O
are	O
.	O
We	O
also	O
mention	O
preliminary	O
results	O
on	O
model	B-KEY
checking	I-KEY
alternating-time	O
temporal	O
logic	O
for	O
this	O
natural	O
class	O
of	O
models	O
.	O

Learning	O
Consumer	B-KEY
Preferences	I-KEY
Using	O
Semantic	B-KEY
Similarity	I-KEY
∗	O
Reyhan	O
Aydo˘gan	O
Pınar	O
Yolum	O
ABSTRACT	O
In	O
online	O
,	O
dynamic	O
environments	O
,	O
the	O
services	B-KEY
requested	O
by	O
consumers	O
may	O
not	O
be	O
readily	O
served	O
by	O
the	O
providers	O
.	O
This	O
requires	O
the	O
service	B-KEY
consumers	O
and	O
providers	O
to	O
negotiate	B-KEY
their	O
service	B-KEY
needs	O
and	O
offers	O
.	O
Multiagent	O
negotiation	B-KEY
approaches	O
typically	O
assume	O
that	O
the	O
parties	O
agree	O
on	O
service	B-KEY
content	O
and	O
focus	O
on	O
finding	O
a	O
consensus	O
on	O
service	B-KEY
price	B-KEY
.	O
In	O
contrast	O
,	O
this	O
work	O
develops	O
an	O
approach	O
through	O
which	O
the	O
parties	O
can	O
negotiate	B-KEY
the	O
content	O
of	O
a	O
service	B-KEY
.	O
This	O
calls	O
for	O
a	O
negotiation	B-KEY
approach	O
in	O
which	O
the	O
parties	O
can	O
understand	O
the	O
semantics	O
of	O
their	O
requests	O
and	O
offers	O
and	O
learn	O
each	O
other	O
's	O
preferences	O
incrementally	O
over	O
time	O
.	O
Accordingly	O
,	O
we	O
propose	O
an	O
architecture	O
in	O
which	O
both	O
consumers	O
and	O
producers	O
use	O
a	O
shared	O
ontology	B-KEY
to	O
negotiate	B-KEY
a	O
service	B-KEY
.	O
Through	O
repetitive	O
interactions	O
,	O
the	O
provider	O
learns	O
consumers	O
'	O
needs	O
accurately	O
and	O
can	O
make	O
better	O
targeted	O
offers	O
.	O
To	O
enable	O
fast	O
and	O
accurate	O
learning	O
of	O
preferences	O
,	O
we	O
develop	O
an	O
extension	O
to	O
Version	O
Space	O
and	O
compare	O
it	O
with	O
existing	O
learning	O
techniques	O
.	O
We	O
further	O
develop	O
a	O
metric	O
for	O
measuring	O
semantic	B-KEY
similarity	I-KEY
between	O
services	B-KEY
and	O
compare	O
the	O
performance	O
of	O
our	O
approach	O
using	O
different	O
similarity	B-KEY
metrics	I-KEY
.	O

The	O
LOGIC	O
Negotiation	B-KEY
Model	O
ABSTRACT	O
Successful	O
negotiators	B-KEY
prepare	O
by	O
determining	O
their	O
position	O
along	O
five	O
dimensions	O
:	O
Legitimacy	O
,	O
Options	O
,	O
Goals	O
,	O
Independence	O
,	O
and	O
Commitment	O
,	O
-LRB-	O
LOGIC	O
-RRB-	O
.	O
We	O
introduce	O
a	O
negotiation	B-KEY
model	O
based	O
on	O
these	O
dimensions	O
and	O
on	O
two	O
primitive	O
concepts	O
:	O
intimacy	O
-LRB-	O
degree	O
of	O
closeness	O
-RRB-	O
and	O
balance	O
-LRB-	O
degree	O
of	O
fairness	O
-RRB-	O
.	O
The	O
intimacy	O
is	O
a	O
pair	O
of	O
matrices	O
that	O
evaluate	O
both	O
an	O
agent	O
's	O
contribution	O
to	O
the	O
relationship	O
and	O
its	O
opponent	O
's	O
contribution	O
each	O
from	O
an	O
information	O
view	O
and	O
from	O
a	O
utilitarian	O
view	O
across	O
the	O
five	O
LOGIC	O
dimensions	O
.	O
The	O
balance	O
is	O
the	O
difference	O
between	O
these	O
matrices	O
.	O
A	O
relationship	O
strategy	O
maintains	O
a	O
target	O
intimacy	O
for	O
each	O
relationship	O
that	O
an	O
agent	O
would	O
like	O
the	O
relationship	O
to	O
move	O
towards	O
in	O
future	O
.	O
The	O
negotiation	B-KEY
strategy	O
maintains	O
a	O
set	O
of	O
Options	O
that	O
are	O
in-line	O
with	O
the	O
current	O
intimacy	O
level	O
,	O
and	O
then	O
tactics	O
wrap	O
the	O
Options	O
in	O
argumentation	O
with	O
the	O
aim	O
of	O
attaining	O
a	O
successful	O
deal	O
and	O
manipulating	O
the	O
successive	O
negotiation	O
balances	O
towards	O
the	O
target	O
intimacy	O
.	O

A	O
Frequency-based	O
and	O
a	O
Poisson-based	O
Definition	O
of	O
the	O
Probability	O
of	O
Being	O
Informative	B-KEY
ABSTRACT	O
This	O
paper	O
reports	O
on	O
theoretical	O
investigations	O
about	O
the	O
assumptions	O
underlying	O
the	O
inverse	B-KEY
document	I-KEY
frequency	I-KEY
-LRB-	O
idf	B-KEY
-RRB-	O
.	O
We	O
show	O
that	O
an	O
intuitive	O
idf	B-KEY
-	O
based	O
probability	B-KEY
function	I-KEY
for	O
the	O
probability	O
of	O
a	O
term	O
being	O
informative	B-KEY
assumes	O
disjoint	O
document	O
events	O
.	O
By	O
assuming	O
documents	O
to	O
be	O
independent	O
rather	O
than	O
disjoint	O
,	O
we	O
arrive	O
at	O
a	O
Poisson-based	O
probability	O
of	O
being	O
informative	B-KEY
.	O
The	O
framework	O
is	O
useful	O
for	O
understanding	O
and	O
deciding	O
the	O
parameter	O
estimation	O
and	O
combination	O
in	O
probabilistic	O
retrieval	O
models	O
.	O

Controlling	O
Overlap	O
in	O
Content-Oriented	O
XML	B-KEY
Retrieval	O
ABSTRACT	O
The	O
direct	O
application	O
of	O
standard	O
ranking	B-KEY
techniques	O
to	O
retrieve	O
individual	O
elements	O
from	O
a	O
collection	O
of	O
XML	B-KEY
documents	O
often	O
produces	O
a	O
result	O
set	O
in	O
which	O
the	O
top	O
ranks	B-KEY
are	O
dominated	O
by	O
a	O
large	O
number	O
of	O
elements	O
taken	O
from	O
a	O
small	O
number	O
of	O
highly	O
relevant	O
documents	O
.	O
This	O
paper	O
presents	O
and	O
evaluates	O
an	O
algorithm	O
that	O
re-ranks	O
this	O
result	O
set	O
,	O
with	O
the	O
aim	O
of	O
minimizing	O
redundant	O
content	O
while	O
preserving	O
the	O
benefits	O
of	O
element	O
retrieval	O
,	O
including	O
the	O
benefit	O
of	O
identifying	O
topic-focused	O
components	O
contained	O
within	O
relevant	O
documents	O
.	O
The	O
test	O
collection	O
developed	O
by	O
the	O
INitiative	O
for	O
the	O
Evaluation	O
of	O
XML	B-KEY
Retrieval	O
-LRB-	O
INEX	B-KEY
-RRB-	O
forms	O
the	O
basis	O
for	O
the	O
evaluation	O
.	O

A	O
Multi-Agent	O
System	O
for	O
Building	O
Dynamic	O
Ontologies	B-KEY
ABSTRACT	O
Ontologies	B-KEY
building	O
from	O
text	O
is	O
still	O
a	O
time-consuming	O
task	O
which	O
justifies	O
the	O
growth	O
of	O
Ontology	B-KEY
Learning	O
.	O
Our	O
system	O
named	O
Dynamo	B-KEY
is	O
designed	O
along	O
this	O
domain	O
but	O
following	O
an	O
original	O
approach	O
based	O
on	O
an	O
adaptive	O
multi-agent	O
architecture	O
.	O
In	O
this	O
paper	O
we	O
present	O
a	O
distributed	O
hierarchical	O
clustering	O
algorithm	O
,	O
core	O
of	O
our	O
approach	O
.	O
It	O
is	O
evaluated	O
and	O
compared	O
to	O
a	O
more	O
conventional	O
centralized	O
algorithm	O
.	O
We	O
also	O
present	O
how	O
it	O
has	O
been	O
improved	O
using	O
a	O
multi-criteria	O
approach	O
.	O
With	O
those	O
results	O
in	O
mind	O
,	O
we	O
discuss	O
the	O
limits	O
of	O
our	O
system	O
and	O
add	O
as	O
perspectives	O
the	O
modifications	O
required	O
to	O
reach	O
a	O
complete	O
ontology	B-KEY
building	O
solution	O
.	O

Remote	B-KEY
Access	I-KEY
to	O
Large	O
Spatial	O
Databases	O
*	O
ABSTRACT	O
Enterprises	O
in	O
the	O
public	O
and	O
private	O
sectors	O
have	O
been	O
making	O
their	O
large	B-KEY
spatial	I-KEY
data	I-KEY
archives	O
available	O
over	O
the	O
Internet	B-KEY
.	O
However	O
,	O
interactive	O
work	O
with	O
such	O
large	O
volumes	O
of	O
online	O
spatial	O
data	O
is	O
a	O
challenging	O
task	O
.	O
We	O
propose	O
two	O
efficient	O
approaches	O
to	O
remote	B-KEY
access	I-KEY
to	O
large	B-KEY
spatial	I-KEY
data	I-KEY
.	O
First	O
,	O
we	O
introduce	O
a	O
client-server	O
architecture	O
where	O
the	O
work	O
is	O
distributed	O
between	O
the	O
server	O
and	O
the	O
individual	O
clients	O
for	O
spatial	B-KEY
query	I-KEY
evaluation	I-KEY
,	O
data	B-KEY
visualization	I-KEY
,	O
and	O
data	B-KEY
management	I-KEY
.	O
We	O
enable	O
the	O
minimization	O
of	O
the	O
requirements	O
for	O
system	O
resources	O
on	O
the	O
client	O
side	O
while	O
maximizing	O
system	O
responsiveness	O
as	O
well	O
as	O
the	O
number	O
of	O
connections	O
one	O
server	O
can	O
handle	O
concurrently	O
.	O
Second	O
,	O
for	O
prolonged	O
periods	O
of	O
access	O
to	O
large	O
online	O
data	O
,	O
we	O
introduce	O
APPOINT	O
-LRB-	O
an	O
Approach	O
for	O
Peer-to-Peer	O
Offloading	O
the	O
INTernet	B-KEY
-RRB-	O
.	O
This	O
is	O
a	O
centralized	O
peer-to-peer	O
approach	O
that	O
helps	O
Internet	B-KEY
users	O
transfer	O
large	O
volumes	O
of	O
online	O
data	O
efficiently	O
.	O
In	O
APPOINT	O
,	O
active	O
clients	O
of	O
the	O
clientserver	O
architecture	O
act	O
on	O
the	O
server	O
's	O
behalf	O
and	O
communicate	O
with	O
each	O
other	O
to	O
decrease	O
network	B-KEY
latency	I-KEY
,	O
improve	O
service	O
bandwidth	O
,	O
and	O
resolve	O
server	O
congestions	O
.	O

Combining	B-KEY
Content	I-KEY
and	I-KEY
Link	I-KEY
for	O
Classification	B-KEY
using	O
Matrix	B-KEY
Factorization	I-KEY
ABSTRACT	O
The	O
world	O
wide	O
web	O
contains	O
rich	O
textual	O
contents	O
that	O
are	O
interconnected	O
via	O
complex	O
hyperlinks	O
.	O
This	O
huge	O
database	O
violates	O
the	O
assumption	O
held	O
by	O
most	O
of	O
conventional	O
statistical	O
methods	O
that	O
each	O
web	O
page	O
is	O
considered	O
as	O
an	O
independent	O
and	O
identical	O
sample	O
.	O
It	O
is	O
thus	O
difficult	O
to	O
apply	O
traditional	O
mining	O
or	O
learning	O
methods	O
for	O
solving	O
web	B-KEY
mining	I-KEY
problems	I-KEY
,	O
e.g.	O
,	O
web	O
page	O
classification	B-KEY
,	O
by	O
exploiting	O
both	O
the	O
content	O
and	O
the	O
link	B-KEY
structure	I-KEY
.	O
The	O
research	O
in	O
this	O
direction	O
has	O
recently	O
received	O
considerable	O
attention	O
but	O
are	O
still	O
in	O
an	O
early	O
stage	O
.	O
Though	O
a	O
few	O
methods	O
exploit	O
both	O
the	O
link	B-KEY
structure	I-KEY
or	O
the	O
content	B-KEY
information	I-KEY
,	O
some	O
of	O
them	O
combine	O
the	O
only	O
authority	B-KEY
information	I-KEY
with	O
the	O
content	B-KEY
information	I-KEY
,	O
and	O
the	O
others	O
first	O
decompose	O
the	O
link	B-KEY
structure	I-KEY
into	O
hub	O
and	O
authority	O
features	O
,	O
then	O
apply	O
them	O
as	O
additional	O
document	O
features	O
.	O
Being	O
practically	O
attractive	O
for	O
its	O
great	O
simplicity	O
,	O
this	O
paper	O
aims	O
to	O
design	O
an	O
algorithm	O
that	O
exploits	O
both	O
the	O
content	O
and	O
linkage	O
information	O
,	O
by	O
carrying	O
out	O
a	O
joint	B-KEY
factorization	I-KEY
on	O
both	O
the	O
linkage	B-KEY
adjacency	I-KEY
matrix	I-KEY
and	O
the	O
document-term	B-KEY
matrix	I-KEY
,	O
and	O
derives	O
a	O
new	O
representation	O
for	O
web	O
pages	O
in	O
a	O
low-dimensional	B-KEY
factor	I-KEY
space	I-KEY
,	O
without	O
explicitly	O
separating	O
them	O
as	O
content	O
,	O
hub	O
or	O
authority	O
factors	O
.	O
Further	O
analysis	O
can	O
be	O
performed	O
based	O
on	O
the	O
compact	O
representation	O
of	O
web	O
pages	O
.	O
In	O
the	O
experiments	O
,	O
the	O
proposed	O
method	O
is	O
compared	O
with	O
state-of-the-art	O
methods	O
and	O
demonstrates	O
an	O
excellent	O
accuracy	O
in	O
hypertext	O
classification	B-KEY
on	O
the	O
WebKB	B-KEY
and	I-KEY
Cora	I-KEY
benchmarks	I-KEY
.	O

Event	B-KEY
Threading	O
within	O
News	O
Topics	O
ABSTRACT	O
With	O
the	O
overwhelming	O
volume	O
of	O
online	O
news	O
available	O
today	O
,	O
there	O
is	O
an	O
increasing	O
need	O
for	O
automatic	B-KEY
techniques	I-KEY
to	O
analyze	O
and	O
present	O
news	O
to	O
the	O
user	O
in	O
a	O
meaningful	O
and	O
efficient	O
manner	O
.	O
Previous	O
research	O
focused	O
only	O
on	O
organizing	O
news	O
stories	O
by	O
their	O
topics	O
into	O
a	O
flat	B-KEY
hierarchy	I-KEY
.	O
We	O
believe	O
viewing	O
a	O
news	O
topic	O
as	O
a	O
flat	O
collection	O
of	O
stories	O
is	O
too	O
restrictive	O
and	O
inefficient	O
for	O
a	O
user	O
to	O
understand	O
the	O
topic	O
quickly	O
.	O
In	O
this	O
work	O
,	O
we	O
attempt	O
to	O
capture	O
the	O
rich	O
structure	O
of	O
events	B-KEY
and	O
their	O
dependencies	B-KEY
in	O
a	O
news	O
topic	O
through	O
our	O
event	B-KEY
models	O
.	O
We	O
call	O
the	O
process	O
of	O
recognizing	O
events	B-KEY
and	O
their	O
dependencies	B-KEY
event	B-KEY
threading	O
.	O
We	O
believe	O
our	O
perspective	O
of	O
modeling	O
the	O
structure	O
of	O
a	O
topic	O
is	O
more	O
effective	O
in	O
capturing	O
its	O
semantics	O
than	O
a	O
flat	O
list	O
of	O
on-topic	O
stories	O
.	O
We	O
formally	O
define	O
the	O
novel	O
problem	O
,	O
suggest	O
evaluation	O
metrics	O
and	O
present	O
a	O
few	O
techniques	O
for	O
solving	O
the	O
problem	O
.	O
Besides	O
the	O
standard	O
word	O
based	O
features	O
,	O
our	O
approaches	O
take	O
into	O
account	O
novel	B-KEY
features	I-KEY
such	O
as	O
temporal	B-KEY
locality	I-KEY
of	O
stories	O
for	O
event	B-KEY
recognition	O
and	O
time-ordering	O
for	O
capturing	O
dependencies	O
.	O
Our	O
experiments	O
on	O
a	O
manually	O
labeled	O
data	O
sets	O
show	O
that	O
our	O
models	O
effectively	O
identify	O
the	O
events	B-KEY
and	O
capture	O
dependencies	B-KEY
among	O
them	O
.	O

A	O
Cross-Layer	O
Approach	O
to	O
Resource	B-KEY
Discovery	I-KEY
and	O
Distribution	O
in	O
Mobile	O
Ad	O
hoc	O
Networks	O
ABSTRACT	O
This	O
paper	O
describes	O
a	O
cross-layer	O
approach	O
to	O
designing	O
robust	O
P2P	O
system	O
over	O
mobile	O
ad	O
hoc	O
networks	O
.	O
The	O
design	O
is	O
based	O
on	O
simple	O
functional	O
primitives	O
that	O
allow	O
routing	O
at	O
both	O
P2P	O
and	O
network	O
layers	O
to	O
be	O
integrated	O
to	O
reduce	O
overhead	O
.	O
With	O
these	O
primitives	O
,	O
the	O
paper	O
addresses	O
various	O
load	O
balancing	O
techniques	O
.	O
Preliminary	O
simulation	O
results	O
are	O
also	O
presented	O
.	O

Concept	O
and	O
Architecture	O
of	O
a	O
Pervasive	B-KEY
Document	I-KEY
Editing	I-KEY
and	I-KEY
Managing	I-KEY
System	I-KEY
ABSTRACT	O
Collaborative	B-KEY
document	I-KEY
processing	O
has	O
been	O
addressed	O
by	O
many	O
approaches	O
so	O
far	O
,	O
most	O
of	O
which	O
focus	O
on	O
document	O
versioning	O
and	O
collaborative	O
editing	O
.	O
We	O
address	O
this	O
issue	O
from	O
a	O
different	O
angle	O
and	O
describe	O
the	O
concept	O
and	O
architecture	O
of	O
a	O
pervasive	B-KEY
document	I-KEY
editing	I-KEY
and	I-KEY
managing	I-KEY
system	I-KEY
.	O
It	O
exploits	O
database	O
techniques	O
and	O
real-time	O
updating	O
for	O
sophisticated	O
collaboration	O
scenarios	O
on	O
multiple	O
devices	O
.	O
Each	O
user	O
is	O
always	O
served	O
with	O
upto-date	O
documents	O
and	O
can	O
organize	O
his	O
work	O
based	O
on	O
document	O
meta	O
data	O
.	O
For	O
this	O
,	O
we	O
present	O
our	O
conceptual	O
architecture	O
for	O
such	O
a	O
system	O
and	O
discuss	O
it	O
with	O
an	O
example	O
.	O

Finding	O
Equilibria	O
in	O
Large	O
Sequential	B-KEY
Games	I-KEY
of	O
Imperfect	B-KEY
Information	I-KEY
*	O
ABSTRACT	O
Finding	O
an	O
equilibrium	B-KEY
of	O
an	O
extensive	O
form	O
game	O
of	O
imperfect	B-KEY
information	I-KEY
is	O
a	O
fundamental	O
problem	O
in	O
computational	B-KEY
game	I-KEY
theory	I-KEY
,	O
but	O
current	O
techniques	O
do	O
not	O
scale	O
to	O
large	O
games	O
.	O
To	O
address	O
this	O
,	O
we	O
introduce	O
the	O
ordered	B-KEY
game	I-KEY
isomorphism	I-KEY
and	O
the	O
related	O
ordered	B-KEY
game	I-KEY
isomorphic	I-KEY
abstraction	O
transformation	O
.	O
For	O
a	O
multi-player	O
sequential	B-KEY
game	I-KEY
of	O
imperfect	B-KEY
information	I-KEY
with	O
observable	O
actions	O
and	O
an	O
ordered	O
signal	O
space	O
,	O
we	O
prove	O
that	O
any	O
Nash	O
equilibrium	O
in	O
an	O
abstracted	O
smaller	O
game	O
,	O
obtained	O
by	O
one	O
or	O
more	O
applications	O
of	O
the	O
transformation	O
,	O
can	O
be	O
easily	O
converted	O
into	O
a	O
Nash	O
equilibrium	O
in	O
the	O
original	O
game	O
.	O
We	O
present	O
an	O
algorithm	O
,	O
GameShrink	B-KEY
,	O
for	O
abstracting	O
the	O
game	O
using	O
our	O
isomorphism	O
exhaustively	O
.	O
Its	O
complexity	O
is	O
˜O	O
-LRB-	O
n2	O
-RRB-	O
,	O
where	O
n	O
is	O
the	O
number	O
of	O
nodes	O
in	O
a	O
structure	O
we	O
call	O
the	O
signal	B-KEY
tree	I-KEY
.	O
It	O
is	O
no	O
larger	O
than	O
the	O
game	O
tree	O
,	O
and	O
on	O
nontrivial	O
games	O
it	O
is	O
drastically	O
smaller	O
,	O
so	O
GameShrink	B-KEY
has	O
time	O
and	O
space	O
complexity	O
sublinear	O
in	O
the	O
size	O
of	O
the	O
game	O
tree	O
.	O
Using	O
GameShrink	B-KEY
,	O
we	O
find	O
an	O
equilibrium	B-KEY
to	O
a	O
poker	O
game	O
with	O
3.1	O
billion	O
nodes	O
--	O
over	O
four	O
orders	O
of	O
magnitude	O
more	O
than	O
in	O
the	O
largest	O
poker	O
game	O
solved	O
previously	O
.	O
We	O
discuss	O
several	O
electronic	O
commerce	O
applications	O
for	O
GameShrink	B-KEY
.	O
To	O
address	O
even	O
larger	O
games	O
,	O
we	O
introduce	O
approximation	O
methods	O
that	O
do	O
not	O
preserve	O
equilibrium	B-KEY
,	O
but	O
nevertheless	O
yield	O
-LRB-	O
ex	O
post	O
-RRB-	O
provably	O
close-to-optimal	O
strategies	O
.	O

On	O
the	O
relevance	B-KEY
of	O
utterances	O
in	O
formal	O
inter-agent	O
dialogues	B-KEY
ABSTRACT	O
Work	O
on	O
argumentation-based	O
dialogue	B-KEY
has	O
defined	O
frameworks	O
within	O
which	O
dialogues	B-KEY
can	O
be	O
carried	O
out	O
,	O
established	O
protocols	O
that	O
govern	O
dialogues	B-KEY
,	O
and	O
studied	O
different	O
properties	O
of	O
dialogues	B-KEY
.	O
This	O
work	O
has	O
established	O
the	O
space	O
in	O
which	O
agents	O
are	O
permitted	O
to	O
interact	O
through	O
dialogues	B-KEY
.	O
Recently	O
,	O
there	O
has	O
been	O
increasing	O
interest	O
in	O
the	O
mechanisms	O
agents	O
might	O
use	O
to	O
choose	O
how	O
to	O
act	O
--	O
the	O
rhetorical	O
manoeuvring	O
that	O
they	O
use	O
to	O
navigate	O
through	O
the	O
space	O
defined	O
by	O
the	O
rules	O
of	O
the	O
dialogue	B-KEY
.	O
Key	O
in	O
such	O
considerations	O
is	O
the	O
idea	O
of	O
relevance	B-KEY
,	O
since	O
a	O
usual	O
requirement	O
is	O
that	O
agents	O
stay	O
focussed	O
on	O
the	O
subject	O
of	O
the	O
dialogue	B-KEY
and	O
only	O
make	O
relevant	B-KEY
remarks	O
.	O
Here	O
we	O
study	O
several	O
notions	O
of	O
relevance	B-KEY
,	O
showing	O
how	O
they	O
can	O
be	O
related	O
to	O
both	O
the	O
rules	O
for	O
carrying	O
out	O
dialogues	B-KEY
and	O
to	O
rhetorical	O
manoeuvring	O
.	O

A	O
Randomized	O
Method	O
for	O
the	O
Shapley	O
Value	O
for	O
the	O
Voting	O
Game	O
ABSTRACT	O
The	O
Shapley	O
value	O
is	O
one	O
of	O
the	O
key	O
solution	O
concepts	O
for	O
coalition	O
games	O
.	O
Its	O
main	O
advantage	O
is	O
that	O
it	O
provides	O
a	O
unique	B-KEY
and	I-KEY
fair	I-KEY
solution	I-KEY
,	O
but	O
its	O
main	O
problem	O
is	O
that	O
,	O
for	O
many	O
coalition	O
games	O
,	O
the	O
Shapley	O
value	O
can	O
not	O
be	O
determined	O
in	O
polynomial	B-KEY
time	I-KEY
.	O
In	O
particular	O
,	O
the	O
problem	O
of	O
finding	O
this	O
value	O
for	O
the	O
voting	O
game	O
is	O
known	O
to	O
be	O
#P	O
-	O
complete	O
in	O
the	O
general	O
case	O
.	O
However	O
,	O
in	O
this	O
paper	O
,	O
we	O
show	O
that	O
there	O
are	O
some	O
specific	O
voting	O
games	O
for	O
which	O
the	O
problem	O
is	O
computationally	O
tractable	O
.	O
For	O
other	O
general	O
voting	O
games	O
,	O
we	O
overcome	O
the	O
problem	O
of	O
computational	O
complexity	O
by	O
presenting	O
a	O
new	O
randomized	O
method	O
for	O
determining	O
the	O
approximate	B-KEY
Shapley	O
value	O
.	O
The	O
time	O
complexity	O
of	O
this	O
method	O
is	O
linear	O
in	O
the	O
number	O
of	O
players	O
.	O
We	O
also	O
show	O
,	O
through	O
empirical	O
studies	O
,	O
that	O
the	O
percentage	O
error	O
for	O
the	O
proposed	O
method	O
is	O
always	O
less	O
than	O
20	O
%	O
and	O
,	O
in	O
most	O
cases	O
,	O
less	O
than	O
5	O
%	O
.	O

Automatic	O
Extraction	O
of	O
Titles	O
from	O
General	O
Documents	O
using	O
Machine	B-KEY
Learning	I-KEY
ABSTRACT	O
In	O
this	O
paper	O
,	O
we	O
propose	O
a	O
machine	B-KEY
learning	I-KEY
approach	O
to	O
title	B-KEY
extraction	I-KEY
from	O
general	O
documents	O
.	O
By	O
general	O
documents	O
,	O
we	O
mean	O
documents	O
that	O
can	O
belong	O
to	O
any	O
one	O
of	O
a	O
number	O
of	O
specific	O
genres	B-KEY
,	O
including	O
presentations	O
,	O
book	O
chapters	O
,	O
technical	O
papers	O
,	O
brochures	O
,	O
reports	O
,	O
and	O
letters	O
.	O
Previously	O
,	O
methods	O
have	O
been	O
proposed	O
mainly	O
for	O
title	B-KEY
extraction	I-KEY
from	O
research	O
papers	O
.	O
It	O
has	O
not	O
been	O
clear	O
whether	O
it	O
could	O
be	O
possible	O
to	O
conduct	O
automatic	B-KEY
title	I-KEY
extraction	I-KEY
from	O
general	O
documents	O
.	O
As	O
a	O
case	O
study	O
,	O
we	O
consider	O
extraction	O
from	O
Office	O
including	O
Word	O
and	O
PowerPoint	O
.	O
In	O
our	O
approach	O
,	O
we	O
annotate	O
titles	O
in	O
sample	O
documents	O
-LRB-	O
for	O
Word	O
and	O
PowerPoint	O
respectively	O
-RRB-	O
and	O
take	O
them	O
as	O
training	O
data	O
,	O
train	O
machine	B-KEY
learning	I-KEY
models	O
,	O
and	O
perform	O
title	B-KEY
extraction	I-KEY
using	O
the	O
trained	O
models	O
.	O
Our	O
method	O
is	O
unique	O
in	O
that	O
we	O
mainly	O
utilize	O
formatting	B-KEY
information	I-KEY
such	O
as	O
font	O
size	O
as	O
features	O
in	O
the	O
models	O
.	O
It	O
turns	O
out	O
that	O
the	O
use	O
of	O
formatting	B-KEY
information	I-KEY
can	O
lead	O
to	O
quite	O
accurate	O
extraction	O
from	O
general	O
documents	O
.	O
Precision	O
and	O
recall	O
for	O
title	B-KEY
extraction	I-KEY
from	O
Word	O
is	O
0.810	O
and	O
0.837	O
respectively	O
,	O
and	O
precision	O
and	O
recall	O
for	O
title	B-KEY
extraction	I-KEY
from	O
PowerPoint	O
is	O
0.875	O
and	O
0.895	O
respectively	O
in	O
an	O
experiment	O
on	O
intranet	O
data	O
.	O
Other	O
important	O
new	O
findings	O
in	O
this	O
work	O
include	O
that	O
we	O
can	O
train	O
models	O
in	O
one	O
domain	O
and	O
apply	O
them	O
to	O
another	O
domain	O
,	O
and	O
more	O
surprisingly	O
we	O
can	O
even	O
train	O
models	O
in	O
one	O
language	O
and	O
apply	O
them	O
to	O
another	O
language	O
.	O
Moreover	O
,	O
we	O
can	O
significantly	O
improve	O
search	B-KEY
ranking	O
results	O
in	O
document	B-KEY
retrieval	I-KEY
by	O
using	O
the	O
extracted	O
titles	O
.	O

Context-Sensitive	O
Information	O
Retrieval	O
Using	O
Implicit	O
Feedback	O
ABSTRACT	O
A	O
major	O
limitation	O
of	O
most	O
existing	O
retrieval	O
models	O
and	O
systems	O
is	O
that	O
the	O
retrieval	O
decision	O
is	O
made	O
based	O
solely	O
on	O
the	O
query	O
and	O
document	O
collection	O
;	O
information	O
about	O
the	O
actual	O
user	O
and	O
search	O
context	B-KEY
is	O
largely	O
ignored	O
.	O
In	O
this	O
paper	O
,	O
we	O
study	O
how	O
to	O
exploit	O
implicit	B-KEY
feedback	I-KEY
information	I-KEY
,	O
including	O
previous	O
queries	O
and	O
clickthrough	B-KEY
information	I-KEY
,	O
to	O
improve	O
retrieval	B-KEY
accuracy	I-KEY
in	O
an	O
interactive	O
information	O
retrieval	O
setting	O
.	O
We	O
propose	O
several	O
contextsensitive	O
retrieval	O
algorithms	O
based	O
on	O
statistical	O
language	O
models	O
to	O
combine	O
the	O
preceding	O
queries	O
and	O
clicked	O
document	O
summaries	O
with	O
the	O
current	B-KEY
query	I-KEY
for	O
better	O
ranking	O
of	O
documents	O
.	O
We	O
use	O
the	O
TREC	O
AP	O
data	O
to	O
create	O
a	O
test	O
collection	O
with	O
search	O
context	B-KEY
information	O
,	O
and	O
quantitatively	O
evaluate	O
our	O
models	O
using	O
this	O
test	O
set	O
.	O
Experiment	O
results	O
show	O
that	O
using	O
implicit	O
feedback	O
,	O
especially	O
the	O
clicked	O
document	O
summaries	O
,	O
can	O
improve	O
retrieval	O
performance	O
substantially	O
.	O

Competitive	B-KEY
Algorithms	I-KEY
for	O
VWAP	B-KEY
and	O
Limit	O
Order	O
Trading	O
ABSTRACT	O
We	O
introduce	O
new	O
online	B-KEY
models	I-KEY
for	O
two	O
important	O
aspects	O
of	O
modern	B-KEY
financial	I-KEY
markets	I-KEY
:	O
Volume	O
Weighted	O
Average	O
Price	O
trading	O
and	O
limit	O
order	O
books	O
.	O
We	O
provide	O
an	O
extensive	O
study	O
of	O
competitive	B-KEY
algorithms	I-KEY
in	O
these	O
models	O
and	O
relate	O
them	O
to	O
earlier	O
online	B-KEY
algorithms	I-KEY
for	O
stock	B-KEY
trading	I-KEY
.	O

A	O
Formal	O
Model	O
for	O
Situated	O
Semantic	B-KEY
Alignment	I-KEY
ABSTRACT	O
Ontology	B-KEY
matching	O
is	O
currently	O
a	O
key	O
technology	O
to	O
achieve	O
the	O
semantic	B-KEY
alignment	I-KEY
of	O
ontological	B-KEY
entities	O
used	O
by	O
knowledge-based	O
applications	O
,	O
and	O
therefore	O
to	O
enable	O
their	O
interoperability	O
in	O
distributed	O
environments	O
such	O
as	O
multiagent	O
systems	O
.	O
Most	O
ontology	B-KEY
matching	O
mechanisms	O
,	O
however	O
,	O
assume	O
matching	O
prior	O
integration	O
and	O
rely	O
on	O
semantics	O
that	O
has	O
been	O
coded	O
a	O
priori	O
in	O
concept	O
hierarchies	O
or	O
external	O
sources	O
.	O
In	O
this	O
paper	O
,	O
we	O
present	O
a	O
formal	O
model	O
for	O
a	O
semantic	B-KEY
alignment	I-KEY
procedure	O
that	O
incrementally	O
aligns	O
differing	O
conceptualisations	O
of	O
two	O
or	O
more	O
agents	O
relative	O
to	O
their	O
respective	O
perception	O
of	O
the	O
environment	O
or	O
domain	O
they	O
are	O
acting	O
in	O
.	O
It	O
hence	O
makes	O
the	O
situation	O
in	O
which	O
the	O
alignment	O
occurs	O
explicit	O
in	O
the	O
model	O
.	O
We	O
resort	O
to	O
Channel	O
Theory	O
to	O
carry	O
out	O
the	O
formalisation	O
.	O

Robustness	B-KEY
of	O
Adaptive	B-KEY
Filtering	I-KEY
Methods	O
In	O
a	O
Cross-benchmark	B-KEY
Evaluation	I-KEY
ABSTRACT	O
This	O
paper	O
reports	O
a	O
cross-benchmark	B-KEY
evaluation	I-KEY
of	O
regularized	B-KEY
logistic	B-KEY
regression	I-KEY
-LRB-	O
LR	B-KEY
-RRB-	O
and	O
incremental	O
Rocchio	B-KEY
for	O
adaptive	B-KEY
filtering	I-KEY
.	O
Using	O
four	O
corpora	O
from	O
the	O
Topic	B-KEY
Detection	I-KEY
and	O
Tracking	O
-LRB-	O
TDT	O
-RRB-	O
forum	O
and	O
the	O
Text	O
Retrieval	O
Conferences	O
-LRB-	O
TREC	O
-RRB-	O
we	O
evaluated	O
these	O
methods	O
with	O
non-stationary	O
topics	O
at	O
various	O
granularity	O
levels	O
,	O
and	O
measured	O
performance	O
with	O
different	O
utility	O
settings	O
.	O
We	O
found	O
that	O
LR	B-KEY
performs	O
strongly	O
and	O
robustly	O
in	O
optimizing	O
T11SU	O
-LRB-	O
a	O
TREC	O
utility	B-KEY
function	I-KEY
-RRB-	O
while	O
Rocchio	B-KEY
is	O
better	O
for	O
optimizing	O
Ctrk	O
-LRB-	O
the	O
TDT	O
tracking	O
cost	O
-RRB-	O
,	O
a	O
high-recall	O
oriented	O
objective	O
function	O
.	O
Using	O
systematic	O
cross-corpus	B-KEY
parameter	I-KEY
optimization	I-KEY
with	O
both	O
methods	O
,	O
we	O
obtained	O
the	O
best	O
results	O
ever	O
reported	O
on	O
TDT5	O
,	O
TREC10	O
and	O
TREC11	O
.	O
Relevance	B-KEY
feedback	I-KEY
on	O
a	O
small	O
portion	O
-LRB-	O
0.05	O
~	O
0.2	O
%	O
-RRB-	O
of	O
the	O
TDT5	O
test	O
documents	O
yielded	O
significant	O
performance	O
improvements	O
,	O
measuring	O
up	O
to	O
a	O
54	O
%	O
reduction	O
in	O
Ctrk	O
and	O
a	O
20.9	O
%	O
increase	O
in	O
T11SU	O
-LRB-	O
with	O
β	O
=	O
0.1	O
-RRB-	O
,	O
compared	O
to	O
the	O
results	O
of	O
the	O
top-performing	O
system	O
in	O
TDT2004	O
without	O
relevance	B-KEY
feedback	I-KEY
information	O
.	O

Operational	B-KEY
Semantics	I-KEY
of	O
Multiagent	B-KEY
Interactions	I-KEY
ABSTRACT	O
The	O
social	O
stance	O
advocated	O
by	O
institutional	B-KEY
frameworks	I-KEY
and	O
most	O
multi-agent	O
system	O
methodologies	O
has	O
resulted	O
in	O
a	O
wide	O
spectrum	O
of	O
organizational	B-KEY
and	I-KEY
communicative	I-KEY
abstractions	I-KEY
which	O
have	O
found	O
currency	O
in	O
several	O
programming	O
frameworks	O
and	O
software	O
platforms	O
.	O
Still	O
,	O
these	O
tools	O
and	O
frameworks	O
are	O
designed	O
to	O
support	O
a	O
limited	O
range	O
of	O
interaction	O
capabilities	O
that	O
constrain	O
developers	O
to	O
a	O
fixed	O
set	O
of	O
particular	O
,	O
pre-defined	B-KEY
abstractions	I-KEY
.	O
The	O
main	O
hypothesis	O
motivating	O
this	O
paper	O
is	O
that	O
the	O
variety	O
of	O
multi-agent	O
interaction	O
mechanisms	O
--	O
both	O
,	O
organizational	O
and	O
communicative	O
,	O
share	O
a	O
common	O
semantic	O
core	O
.	O
In	O
the	O
realm	O
of	O
software	B-KEY
architectures	I-KEY
,	O
the	O
paper	O
proposes	O
a	O
connector-based	O
model	O
of	O
multi-agent	O
interactions	O
which	O
attempts	O
to	O
identify	O
the	O
essential	O
structure	O
underlying	O
multi-agent	O
interactions	O
.	O
Furthermore	O
,	O
the	O
paper	O
also	O
provides	O
this	O
model	O
with	O
a	O
formal	B-KEY
execution	I-KEY
semantics	I-KEY
which	O
describes	O
the	O
dynamics	O
of	O
social	B-KEY
interactions	I-KEY
.	O
The	O
proposed	O
model	O
is	O
intended	O
as	O
the	O
abstract	O
machine	O
of	O
an	O
organizational	B-KEY
programming	I-KEY
language	I-KEY
which	O
allows	O
programmers	O
to	O
accommodate	O
an	O
open	O
set	O
of	O
interaction	O
mechanisms	O
.	O

Evaluating	O
Opportunistic	O
Routing	B-KEY
Protocols	O
with	O
Large	O
Realistic	O
Contact	O
Traces	O
ABSTRACT	O
Traditional	O
mobile	O
ad	O
hoc	O
network	O
-LRB-	O
MANET	O
-RRB-	O
routing	B-KEY
protocols	O
assume	O
that	O
contemporaneous	O
end-to-end	O
communication	O
paths	O
exist	O
between	O
data	O
senders	O
and	O
receivers	O
.	O
In	O
some	O
mobile	O
ad	O
hoc	O
networks	O
with	O
a	O
sparse	O
node	O
population	O
,	O
an	O
end-to-end	O
communication	O
path	O
may	O
break	O
frequently	O
or	O
may	O
not	O
exist	O
at	O
any	O
time	O
.	O
Many	O
routing	B-KEY
protocols	O
have	O
been	O
proposed	O
in	O
the	O
literature	O
to	O
address	O
the	O
problem	O
,	O
but	O
few	O
were	O
evaluated	O
in	O
a	O
realistic	O
``	O
opportunistic	O
''	O
network	O
setting	O
.	O
We	O
use	O
simulation	B-KEY
and	O
contact	B-KEY
traces	I-KEY
-LRB-	O
derived	O
from	O
logs	O
in	O
a	O
production	O
network	O
-RRB-	O
to	O
evaluate	O
and	O
compare	O
five	O
existing	O
protocols	O
:	O
direct-delivery	O
,	O
epidemic	O
,	O
random	O
,	O
PRoPHET	B-KEY
,	O
and	O
Link-State	O
,	O
as	O
well	O
as	O
our	O
own	O
proposed	O
routing	B-KEY
protocol	O
.	O
We	O
show	O
that	O
the	O
direct	O
delivery	O
and	O
epidemic	O
routing	B-KEY
protocols	O
suffer	O
either	O
low	O
delivery	O
ratio	O
or	O
high	O
resource	O
usage	O
,	O
and	O
other	O
protocols	O
make	O
tradeoffs	O
between	O
delivery	O
ratio	O
and	O
resource	O
usage	O
.	O

AdaRank	O
:	O
A	O
Boosting	B-KEY
Algorithm	O
for	O
Information	B-KEY
Retrieval	I-KEY
ABSTRACT	O
In	O
this	O
paper	O
we	O
address	O
the	O
issue	O
of	O
learning	B-KEY
to	I-KEY
rank	I-KEY
for	O
document	B-KEY
retrieval	I-KEY
.	O
In	O
the	O
task	O
,	O
a	O
model	O
is	O
automatically	O
created	O
with	O
some	O
training	O
data	O
and	O
then	O
is	O
utilized	O
for	O
ranking	O
of	O
documents	O
.	O
The	O
goodness	O
of	O
a	O
model	O
is	O
usually	O
evaluated	O
with	O
performance	O
measures	O
such	O
as	O
MAP	O
-LRB-	O
Mean	O
Average	O
Precision	O
-RRB-	O
and	O
NDCG	O
-LRB-	O
Normalized	O
Discounted	O
Cumulative	O
Gain	O
-RRB-	O
.	O
Ideally	O
a	O
learning	O
algorithm	O
would	O
train	O
a	O
ranking	B-KEY
model	I-KEY
that	O
could	O
directly	O
optimize	O
the	O
performance	O
measures	O
with	O
respect	O
to	O
the	O
training	O
data	O
.	O
Existing	O
methods	O
,	O
however	O
,	O
are	O
only	O
able	O
to	O
train	B-KEY
ranking	I-KEY
models	I-KEY
by	O
minimizing	O
loss	O
functions	O
loosely	O
related	O
to	O
the	O
performance	O
measures	O
.	O
For	O
example	O
,	O
Ranking	O
SVM	O
and	O
RankBoost	B-KEY
train	B-KEY
ranking	I-KEY
models	I-KEY
by	O
minimizing	O
classification	O
errors	O
on	O
instance	O
pairs	O
.	O
To	O
deal	O
with	O
the	O
problem	O
,	O
we	O
propose	O
a	O
novel	B-KEY
learning	I-KEY
algorithm	I-KEY
within	O
the	O
framework	O
of	O
boosting	B-KEY
,	O
which	O
can	O
minimize	O
a	O
loss	O
function	O
directly	O
defined	O
on	O
the	O
performance	O
measures	O
.	O
Our	O
algorithm	O
,	O
referred	O
to	O
as	O
AdaRank	O
,	O
repeatedly	O
constructs	O
`	O
weak	B-KEY
rankers	I-KEY
'	O
on	O
the	O
basis	O
of	O
re-weighted	B-KEY
training	I-KEY
data	I-KEY
and	O
finally	O
linearly	O
combines	O
the	O
weak	B-KEY
rankers	I-KEY
for	O
making	O
ranking	O
predictions	O
.	O
We	O
prove	O
that	O
the	O
training	B-KEY
process	I-KEY
of	O
AdaRank	O
is	O
exactly	O
that	O
of	O
enhancing	O
the	O
performance	O
measure	O
used	O
.	O
Experimental	O
results	O
on	O
four	O
benchmark	O
datasets	O
show	O
that	O
AdaRank	O
significantly	O
outperforms	O
the	O
baseline	O
methods	O
of	O
BM25	O
,	O
Ranking	O
SVM	O
,	O
and	O
RankBoost	B-KEY
.	O

Knowledge-intensive	O
Conceptual	O
Retrieval	O
and	O
Passage	B-KEY
Extraction	I-KEY
of	O
Biomedical	O
Literature	O
ABSTRACT	O
This	O
paper	O
presents	O
a	O
study	O
of	O
incorporating	O
domain-specific	B-KEY
knowledge	I-KEY
-LRB-	O
i.e.	O
,	O
information	O
about	O
concepts	O
and	O
relationships	O
between	O
concepts	O
in	O
a	O
certain	O
domain	O
-RRB-	O
in	O
an	O
information	O
retrieval	O
-LRB-	O
IR	O
-RRB-	O
system	O
to	O
improve	O
its	O
effectiveness	O
in	O
retrieving	O
biomedical	O
literature	O
.	O
The	O
effects	O
of	O
different	O
types	O
of	O
domain-specific	B-KEY
knowledge	I-KEY
in	O
performance	O
contribution	O
are	O
examined	O
.	O
Based	O
on	O
the	O
TREC	O
platform	O
,	O
we	O
show	O
that	O
appropriate	O
use	O
of	O
domainspecific	O
knowledge	O
in	O
a	O
proposed	O
conceptual	O
retrieval	B-KEY
model	I-KEY
yields	O
about	O
23	O
%	O
improvement	O
over	O
the	O
best	O
reported	O
result	O
in	O
passage	O
retrieval	O
in	O
the	O
Genomics	O
Track	O
of	O
TREC	O
2006	O
.	O

On	O
Cheating	B-KEY
in	O
Sealed-Bid	O
Auctions	B-KEY
Motivated	O
by	O
the	O
rise	O
of	O
online	O
auctions	B-KEY
and	O
their	O
relative	O
lack	O
of	O
security	O
,	O
this	O
paper	O
analyzes	O
two	O
forms	O
of	O
cheating	B-KEY
in	O
sealed-bid	O
auctions	B-KEY
.	O
The	O
first	O
type	O
of	O
cheating	B-KEY
we	O
consider	O
occurs	O
when	O
the	O
seller	B-KEY
spies	O
on	O
the	O
bids	O
of	O
a	O
second-price	O
auction	B-KEY
and	O
then	O
inserts	O
a	O
fake	O
bid	O
in	O
order	O
to	O
increase	O
the	O
payment	B-KEY
of	O
the	O
winning	O
bidder	O
.	O
In	O
the	O
second	O
type	O
,	O
a	O
bidder	O
cheats	B-KEY
in	O
a	O
first-price	O
auction	B-KEY
by	O
examining	O
the	O
competing	O
bids	O
before	O
deciding	O
on	O
his	O
own	O
bid	O
.	O
In	O
both	O
cases	B-KEY
,	O
we	O
derive	O
equilibrium	O
strategies	O
when	O
bidders	O
are	O
aware	O
of	O
the	O
possibility	B-KEY
of	O
cheating	B-KEY
.	O
These	O
results	O
provide	O
insights	O
into	O
sealedbid	O
auctions	B-KEY
even	O
in	O
the	O
absence	O
of	O
cheating	B-KEY
,	O
including	O
some	O
counterintuitive	O
results	O
on	O
the	O
effects	O
of	O
overbidding	O
in	O
a	O
first-price	O
auction	B-KEY
.	O

Selfish	O
Caching	B-KEY
in	O
Distributed	B-KEY
Systems	I-KEY
:	O
A	O
Game-Theoretic	O
Analysis	O
ABSTRACT	O
We	O
analyze	O
replication	O
of	O
resources	O
by	O
server	O
nodes	O
that	O
act	O
selfishly	O
,	O
using	O
a	O
game-theoretic	B-KEY
approach	I-KEY
.	O
We	O
refer	O
to	O
this	O
as	O
the	O
selfish	O
caching	B-KEY
problem	O
.	O
In	O
our	O
model	O
,	O
nodes	O
incur	O
either	O
cost	O
for	O
replicating	O
resources	O
or	O
cost	O
for	O
access	O
to	O
a	O
remote	B-KEY
replica	I-KEY
.	O
We	O
show	O
the	O
existence	O
of	O
pure	O
strategy	O
Nash	O
equilibria	O
and	O
investigate	O
the	O
price	B-KEY
of	I-KEY
anarchy	I-KEY
,	O
which	O
is	O
the	O
relative	O
cost	O
of	O
the	O
lack	O
of	O
coordination	O
.	O
The	O
price	B-KEY
of	I-KEY
anarchy	I-KEY
can	O
be	O
high	O
due	O
to	O
undersupply	O
problems	O
,	O
but	O
with	O
certain	O
network	B-KEY
topologies	I-KEY
it	O
has	O
better	O
bounds	O
.	O
With	O
a	O
payment	O
scheme	O
the	O
game	O
can	O
always	O
implement	O
the	O
social	O
optimum	O
in	O
the	O
best	O
case	O
by	O
giving	O
servers	O
incentive	O
to	O
replicate	O
.	O

On	O
Opportunistic	O
Techniques	O
for	O
Solving	O
Decentralized	B-KEY
Markov	I-KEY
Decision	I-KEY
Processes	I-KEY
with	O
Temporal	B-KEY
Constraints	I-KEY
ABSTRACT	O
Decentralized	B-KEY
Markov	I-KEY
Decision	I-KEY
Processes	I-KEY
-LRB-	O
DEC-MDPs	O
-RRB-	O
are	O
a	O
popular	O
model	O
of	O
agent-coordination	B-KEY
problems	I-KEY
in	O
domains	O
with	O
uncertainty	O
and	O
time	O
constraints	O
but	O
very	O
difficult	O
to	O
solve	O
.	O
In	O
this	O
paper	O
,	O
we	O
improve	O
a	O
state-of-the-art	O
heuristic	O
solution	O
method	O
for	O
DEC-MDPs	O
,	O
called	O
OC-DEC-MDP	O
,	O
that	O
has	O
recently	O
been	O
shown	O
to	O
scale	O
up	O
to	O
larger	O
DEC-MDPs	O
.	O
Our	O
heuristic	O
solution	O
method	O
,	O
called	O
Value	B-KEY
Function	I-KEY
Propagation	I-KEY
-LRB-	O
VFP	O
-RRB-	O
,	O
combines	O
two	O
orthogonal	O
improvements	O
of	O
OC-DEC-MDP	O
.	O
First	O
,	O
it	O
speeds	O
up	O
OC-DECMDP	O
by	O
an	O
order	O
of	O
magnitude	O
by	O
maintaining	O
and	O
manipulating	O
a	O
value	O
function	O
for	O
each	O
state	O
-LRB-	O
as	O
a	O
function	O
of	O
time	O
-RRB-	O
rather	O
than	O
a	O
separate	O
value	O
for	O
each	O
pair	O
of	O
sate	O
and	O
time	O
interval	O
.	O
Furthermore	O
,	O
it	O
achieves	O
better	O
solution	O
qualities	O
than	O
OC-DEC-MDP	O
because	O
,	O
as	O
our	O
analytical	O
results	O
show	O
,	O
it	O
does	O
not	O
overestimate	O
the	O
expected	O
total	O
reward	O
like	O
OC-DEC	O
-	O
MDP	O
.	O
We	O
test	O
both	O
improvements	O
independently	O
in	O
a	O
crisis-management	O
domain	O
as	O
well	O
as	O
for	O
other	O
types	O
of	O
domains	O
.	O
Our	O
experimental	O
results	O
demonstrate	O
a	O
significant	O
speedup	O
of	O
VFP	O
over	O
OC-DEC-MDP	O
as	O
well	O
as	O
higher	O
solution	O
qualities	O
in	O
a	O
variety	O
of	O
situations	O
.	O

Approximate	B-KEY
and	O
Online	O
Multi-Issue	O
Negotiation	B-KEY
ABSTRACT	O
This	O
paper	O
analyzes	O
bilateral	O
multi-issue	O
negotiation	B-KEY
between	O
selfinterested	O
autonomous	O
agents	O
.	O
The	O
agents	O
have	O
time	B-KEY
constraints	I-KEY
in	O
the	O
form	O
of	O
both	O
deadlines	O
and	O
discount	O
factors	O
.	O
There	O
are	O
m	O
>	O
1	O
issues	O
for	O
negotiation	B-KEY
where	O
each	O
issue	O
is	O
viewed	O
as	O
a	O
pie	O
of	O
size	O
one	O
.	O
The	O
issues	O
are	O
``	O
indivisible	O
''	O
-LRB-	O
i.e.	O
,	O
individual	O
issues	O
can	O
not	O
be	O
split	O
between	O
the	O
parties	O
;	O
each	O
issue	O
must	O
be	O
allocated	O
in	O
its	O
entirety	O
to	O
either	O
agent	O
-RRB-	O
.	O
Here	O
different	O
agents	O
value	O
different	O
issues	O
differently	O
.	O
Thus	O
,	O
the	O
problem	O
is	O
for	O
the	O
agents	O
to	O
decide	O
how	O
to	O
allocate	O
the	O
issues	O
between	O
themselves	O
so	O
as	O
to	O
maximize	O
their	O
individual	O
utilities	O
.	O
For	O
such	O
negotiations	B-KEY
,	O
we	O
first	O
obtain	O
the	O
equilibrium	B-KEY
strategies	B-KEY
for	O
the	O
case	O
where	O
the	O
issues	O
for	O
negotiation	B-KEY
are	O
known	O
a	O
priori	O
to	O
the	O
parties	O
.	O
Then	O
,	O
we	O
analyse	O
their	O
time	O
complexity	O
and	O
show	O
that	O
finding	O
the	O
equilibrium	B-KEY
offers	O
is	O
an	O
NP-hard	O
problem	O
,	O
even	O
in	O
a	O
complete	O
information	O
setting	O
.	O
In	O
order	O
to	O
overcome	O
this	O
computational	O
complexity	O
,	O
we	O
then	O
present	O
negotiation	B-KEY
strategies	B-KEY
that	O
are	O
approximately	B-KEY
optimal	O
but	O
computationally	O
efficient	O
,	O
and	O
show	O
that	O
they	O
form	O
an	O
equilibrium	B-KEY
.	O
We	O
also	O
analyze	O
the	O
relative	B-KEY
error	I-KEY
-LRB-	O
i.e.	O
,	O
the	O
difference	O
between	O
the	O
true	O
optimum	O
and	O
the	O
approximate	B-KEY
-RRB-	O
.	O
The	O
time	O
complexity	O
of	O
the	O
approximate	B-KEY
equilibrium	B-KEY
strategies	B-KEY
is	O
O	O
-LRB-	O
nm	O
/	O
~	O
2	O
-RRB-	O
where	O
n	O
is	O
the	O
negotiation	B-KEY
deadline	O
and	O
~	O
the	O
relative	B-KEY
error	I-KEY
.	O
Finally	O
,	O
we	O
extend	O
the	O
analysis	O
to	O
online	O
negotiation	B-KEY
where	O
different	O
issues	O
become	O
available	O
at	O
different	O
time	O
points	O
and	O
the	O
agents	O
are	O
uncertain	O
about	O
their	O
valuations	O
for	O
these	O
issues	O
.	O
Specifically	O
,	O
we	O
show	O
that	O
an	O
approximate	B-KEY
equilibrium	B-KEY
exists	O
for	O
online	O
negotiation	B-KEY
and	O
show	O
that	O
the	O
expected	O
difference	O
between	O
the	O
optimum	O
and	O
the	O
approximate	B-KEY
is	O
O	O
-LRB-	O
√	O
m	O
-RRB-	O
.	O
These	O
approximate	B-KEY
strategies	B-KEY
also	O
have	O
polynomial	O
time	O
complexity	O
.	O

CenWits	O
:	O
A	O
Sensor-Based	O
Loosely	O
Coupled	O
Search	B-KEY
and	I-KEY
Rescue	I-KEY
System	O
Using	O
Witnesses	B-KEY
University	O
of	O
Colorado	O
,	O
Campus	O
Box	O
0430	O
Boulder	O
,	O
CO	O
80309-0430	O
ABSTRACT	O
This	O
paper	O
describes	O
the	O
design	O
,	O
implementation	O
and	O
evaluation	O
of	O
a	O
search	B-KEY
and	I-KEY
rescue	I-KEY
system	O
called	O
CenWits	O
.	O
CenWits	O
uses	O
several	O
small	O
,	O
commonly-available	O
RF-based	O
sensors	O
,	O
and	O
a	O
small	O
number	O
of	O
storage	O
and	O
processing	O
devices	O
.	O
It	O
is	O
designed	O
for	O
search	B-KEY
and	I-KEY
rescue	I-KEY
of	O
people	O
in	O
emergency	B-KEY
situations	I-KEY
in	O
wilderness	O
areas	O
.	O
A	O
key	O
feature	O
of	O
CenWits	O
is	O
that	O
it	O
does	O
not	O
require	O
a	O
continuously	O
connected	O
sensor	B-KEY
network	I-KEY
for	O
its	O
operation	O
.	O
It	O
is	O
designed	O
for	O
an	O
intermittently	O
connected	B-KEY
network	I-KEY
that	O
provides	O
only	O
occasional	O
connectivity	O
.	O
It	O
makes	O
a	O
judicious	O
use	O
of	O
the	O
combined	O
storage	O
capability	O
of	O
sensors	O
to	O
filter	O
,	O
organize	O
and	O
store	O
important	O
information	O
,	O
combined	O
battery	O
power	O
of	O
sensors	O
to	O
ensure	O
that	O
the	O
system	O
remains	O
operational	O
for	O
longer	O
time	O
periods	O
,	O
and	O
intermittent	B-KEY
network	I-KEY
connectivity	I-KEY
to	O
propagate	O
information	O
to	O
a	O
processing	O
center	O
.	O
A	O
prototype	O
of	O
CenWits	O
has	O
been	O
implemented	O
using	O
Berkeley	O
Mica2	O
motes	O
.	O
The	O
paper	O
describes	O
this	O
implementation	O
and	O
reports	O
on	O
the	O
performance	O
measured	O
from	O
it	O
.	O

Handling	O
Locations	O
in	O
Search	B-KEY
Engine	I-KEY
Queries	I-KEY
ABSTRACT	O
This	O
paper	O
proposes	O
simple	O
techniques	O
for	O
handling	O
place	B-KEY
references	I-KEY
in	O
search	B-KEY
engine	I-KEY
queries	I-KEY
,	O
an	O
important	O
aspect	O
of	O
geographical	B-KEY
information	I-KEY
retrieval	I-KEY
.	O
We	O
address	O
not	O
only	O
the	O
detection	O
,	O
but	O
also	O
the	O
disambiguation	O
of	O
place	B-KEY
references	I-KEY
,	O
by	O
matching	O
them	O
explicitly	O
with	O
concepts	O
at	O
an	O
ontology	O
.	O
Moreover	O
,	O
when	O
a	O
query	O
does	O
not	O
reference	O
any	O
locations	O
,	O
we	O
propose	O
to	O
use	O
information	O
from	O
documents	O
matching	O
the	O
query	O
,	O
exploiting	O
geographic	O
scopes	O
previously	O
assigned	O
to	O
these	O
documents	O
.	O
Evaluation	O
experiments	O
,	O
using	O
topics	O
from	O
CLEF	O
campaigns	O
and	O
logs	O
from	O
real	O
search	B-KEY
engine	I-KEY
queries	I-KEY
,	O
show	O
the	O
effectiveness	O
of	O
the	O
proposed	O
approaches	O
.	O

GUESS	O
:	O
Gossiping	O
Updates	O
for	O
Efficient	O
Spectrum	B-KEY
Sensing	I-KEY
ABSTRACT	O
Wireless	O
radios	O
of	O
the	O
future	O
will	O
likely	O
be	O
frequency-agile	O
,	O
that	O
is	O
,	O
supporting	O
opportunistic	O
and	O
adaptive	O
use	O
of	O
the	O
RF	B-KEY
spectrum	I-KEY
.	O
Such	O
radios	O
must	O
coordinate	O
with	O
each	O
other	O
to	O
build	O
an	O
accurate	O
and	O
consistent	O
map	O
of	O
spectral	O
utilization	O
in	O
their	O
surroundings	O
.	O
We	O
focus	O
on	O
the	O
problem	O
of	O
sharing	O
RF	B-KEY
spectrum	I-KEY
data	O
among	O
a	O
collection	O
of	O
wireless	O
devices	O
.	O
The	O
inherent	O
requirements	O
of	O
such	O
data	O
and	O
the	O
time-granularity	O
at	O
which	O
it	O
must	O
be	O
collected	O
makes	O
this	O
problem	O
both	O
interesting	O
and	O
technically	O
challenging	O
.	O
We	O
propose	O
GUESS	O
,	O
a	O
novel	O
incremental	O
gossiping	O
approach	O
to	O
coordinated	O
spectral	O
sensing	O
.	O
It	O
-LRB-	O
1	O
-RRB-	O
reduces	O
protocol	O
overhead	O
by	O
limiting	O
the	O
amount	O
of	O
information	O
exchanged	O
between	O
participating	O
nodes	O
,	O
-LRB-	O
2	O
-RRB-	O
is	O
resilient	O
to	O
network	O
alterations	O
,	O
due	O
to	O
node	O
movement	O
or	O
node	O
failures	O
,	O
and	O
-LRB-	O
3	O
-RRB-	O
allows	O
exponentially-fast	O
information	O
convergence	O
.	O
We	O
outline	O
an	O
initial	O
solution	O
incorporating	O
these	O
ideas	O
and	O
also	O
show	O
how	O
our	O
approach	O
reduces	O
network	O
overhead	O
by	O
up	O
to	O
a	O
factor	O
of	O
2.4	O
and	O
results	O
in	O
up	O
to	O
2.7	O
times	O
faster	O
information	O
convergence	O
than	O
alternative	O
approaches	O
.	O

Heuristics-Based	O
Scheduling	B-KEY
of	O
Composite	O
Web	B-KEY
Service	I-KEY
Workloads	O
ABSTRACT	O
Web	B-KEY
services	I-KEY
can	O
be	O
aggregated	O
to	O
create	O
composite	O
workflows	O
that	O
provide	O
streamlined	B-KEY
functionality	I-KEY
for	O
human	O
users	O
or	O
other	O
systems	O
.	O
Although	O
industry	O
standards	O
and	O
recent	O
research	O
have	O
sought	O
to	O
define	O
best	O
practices	O
and	O
to	O
improve	O
end-to-end	B-KEY
workflow	I-KEY
composition	I-KEY
,	O
one	O
area	O
that	O
has	O
not	O
fully	O
been	O
explored	O
is	O
the	O
scheduling	B-KEY
of	O
a	O
workflow	O
's	O
web	B-KEY
service	I-KEY
requests	O
to	O
actual	O
service	O
provisioning	O
in	O
a	O
multi-tiered	O
,	O
multi-organisation	O
environment	O
.	O
This	O
issue	O
is	O
relevant	O
to	O
modern	O
business	O
scenarios	O
where	O
business	O
processes	O
within	O
a	O
workflow	O
must	O
complete	O
within	O
QoS-defined	O
limits	O
.	O
Because	O
these	O
business	O
processes	O
are	O
web	B-KEY
service	I-KEY
consumers	O
,	O
service	B-KEY
requests	I-KEY
must	O
be	O
mapped	O
and	O
scheduled	B-KEY
across	O
multiple	O
web	B-KEY
service	I-KEY
providers	O
,	O
each	O
with	O
its	O
own	O
negotiated	O
service	O
level	O
agreement	O
.	O
In	O
this	O
paper	O
we	O
provide	O
heuristics	B-KEY
for	O
scheduling	B-KEY
service	O
requests	O
from	O
multiple	O
business	O
process	O
workflows	O
to	O
web	O
service	O
providers	O
such	O
that	O
a	O
business	O
value	O
metric	O
across	O
all	O
workflows	O
is	O
maximised	O
.	O
We	O
show	O
that	O
a	O
genetic	O
search	O
algorithm	O
is	O
appropriate	O
to	O
perform	O
this	O
scheduling	B-KEY
,	O
and	O
through	O
experimentation	O
we	O
show	O
that	O
our	O
algorithm	O
scales	O
well	O
up	O
to	O
a	O
thousand	O
workflows	O
and	O
produces	O
better	O
mappings	O
than	O
traditional	O
approaches	O
.	O

Implicit	B-KEY
User	I-KEY
Modeling	I-KEY
for	O
Personalized	O
Search	O
ABSTRACT	O
Information	B-KEY
retrieval	I-KEY
systems	I-KEY
-LRB-	O
e.g.	O
,	O
web	O
search	O
engines	O
-RRB-	O
are	O
critical	O
for	O
overcoming	O
information	O
overload	O
.	O
A	O
major	O
deficiency	O
of	O
existing	O
retrieval	O
systems	O
is	O
that	O
they	O
generally	O
lack	O
user	B-KEY
modeling	I-KEY
and	O
are	O
not	O
adaptive	O
to	O
individual	O
users	O
,	O
resulting	O
in	O
inherently	O
non-optimal	O
retrieval	B-KEY
performance	I-KEY
.	O
For	O
example	O
,	O
a	O
tourist	O
and	O
a	O
programmer	O
may	O
use	O
the	O
same	O
word	O
``	O
java	O
''	O
to	O
search	O
for	O
different	O
information	O
,	O
but	O
the	O
current	O
search	O
systems	O
would	O
return	O
the	O
same	O
results	O
.	O
In	O
this	O
paper	O
,	O
we	O
study	O
how	O
to	O
infer	O
a	O
user	O
's	O
interest	O
from	O
the	O
user	O
's	O
search	O
context	O
and	O
use	O
the	O
inferred	O
implicit	B-KEY
user	I-KEY
model	I-KEY
for	O
personalized	O
search	O
.	O
We	O
present	O
a	O
decision	O
theoretic	O
framework	O
and	O
develop	O
techniques	O
for	O
implicit	B-KEY
user	I-KEY
modeling	I-KEY
in	O
information	O
retrieval	O
.	O
We	O
develop	O
an	O
intelligent	O
client-side	O
web	O
search	O
agent	O
-LRB-	O
UCAIR	O
-RRB-	O
that	O
can	O
perform	O
eager	O
implicit	B-KEY
feedback	I-KEY
,	O
e.g.	O
,	O
query	B-KEY
expansion	I-KEY
based	O
on	O
previous	O
queries	O
and	O
immediate	O
result	O
reranking	O
based	O
on	O
clickthrough	O
information	O
.	O
Experiments	O
on	O
web	O
search	O
show	O
that	O
our	O
search	O
agent	O
can	O
improve	O
search	B-KEY
accuracy	I-KEY
over	O
the	O
popular	O
Google	O
search	O
engine	O
.	O

Composition	O
of	O
a	O
DIDS	O
by	O
Integrating	O
Heterogeneous	O
IDSs	O
on	O
Grids	B-KEY
ABSTRACT	O
This	O
paper	O
considers	O
the	O
composition	O
of	O
a	O
DIDS	O
-LRB-	O
Distributed	B-KEY
Intrusion	I-KEY
Detection	I-KEY
System	I-KEY
-RRB-	O
by	O
integrating	O
heterogeneous	O
IDSs	O
-LRB-	O
Intrusion	O
Detection	O
Systems	O
-RRB-	O
.	O
A	O
Grid	B-KEY
middleware	O
is	O
used	O
for	O
this	O
integration	O
.	O
In	O
addition	O
,	O
an	O
architecture	O
for	O
this	O
integration	O
is	O
proposed	O
and	O
validated	O
through	O
simulation	O
.	O

Distributed	O
Agent-Based	O
Air	O
Traffic	B-KEY
Flow	I-KEY
Management	O
ABSTRACT	O
Air	O
traffic	B-KEY
flow	I-KEY
management	O
is	O
one	O
of	O
the	O
fundamental	O
challenges	O
facing	O
the	O
Federal	O
Aviation	O
Administration	O
-LRB-	O
FAA	O
-RRB-	O
today	O
.	O
The	O
FAA	O
estimates	O
that	O
in	O
2005	O
alone	O
,	O
there	O
were	O
over	O
322,000	O
hours	O
of	O
delays	O
at	O
a	O
cost	O
to	O
the	O
industry	O
in	O
excess	O
of	O
three	O
billion	O
dollars	O
.	O
Finding	O
reliable	O
and	O
adaptive	O
solutions	O
to	O
the	O
flow	O
management	O
problem	O
is	O
of	O
paramount	O
importance	O
if	O
the	O
Next	O
Generation	O
Air	O
Transportation	O
Systems	O
are	O
to	O
achieve	O
the	O
stated	O
goal	O
of	O
accommodating	O
three	O
times	O
the	O
current	O
traffic	O
volume	O
.	O
This	O
problem	O
is	O
particularly	O
complex	O
as	O
it	O
requires	O
the	O
integration	O
and/or	O
coordination	O
of	O
many	O
factors	O
including	O
:	O
new	O
data	O
-LRB-	O
e.g.	O
,	O
changing	O
weather	O
info	O
-RRB-	O
,	O
potentially	O
conflicting	O
priorities	O
-LRB-	O
e.g.	O
,	O
different	O
airlines	O
-RRB-	O
,	O
limited	O
resources	O
-LRB-	O
e.g.	O
,	O
air	B-KEY
traffic	I-KEY
controllers	I-KEY
-RRB-	O
and	O
very	O
heavy	O
traffic	O
volume	O
-LRB-	O
e.g.	O
,	O
over	O
40,000	O
flights	O
over	O
the	O
US	O
airspace	O
-RRB-	O
.	O
In	O
this	O
paper	O
we	O
use	O
FACET	O
--	O
an	O
air	O
traffic	B-KEY
flow	I-KEY
simulator	O
developed	O
at	O
NASA	O
and	O
used	O
extensively	O
by	O
the	O
FAA	O
and	O
industry	O
--	O
to	O
test	O
a	O
multi-agent	O
algorithm	O
for	O
traffic	B-KEY
flow	I-KEY
management	O
.	O
An	O
agent	O
is	O
associated	O
with	O
a	O
fix	O
-LRB-	O
a	O
specific	O
location	O
in	O
2D	O
space	O
-RRB-	O
and	O
its	O
action	O
consists	O
of	O
setting	O
the	O
separation	O
required	O
among	O
the	O
airplanes	O
going	O
though	O
that	O
fix	O
.	O
Agents	O
use	O
reinforcement	B-KEY
learning	I-KEY
to	O
set	O
this	O
separation	O
and	O
their	O
actions	O
speed	O
up	O
or	O
slow	O
down	O
traffic	O
to	O
manage	O
congestion	B-KEY
.	O
Our	O
FACET	O
based	O
results	O
show	O
that	O
agents	O
receiving	O
personalized	O
rewards	O
reduce	O
congestion	B-KEY
by	O
up	O
to	O
45	O
%	O
over	O
agents	O
receiving	O
a	O
global	O
reward	O
and	O
by	O
up	O
to	O
67	O
%	O
over	O
a	O
current	O
industry	O
approach	O
-LRB-	O
Monte	O
Carlo	O
estimation	O
-RRB-	O
.	O

An	O
Efficient	O
Heuristic	B-KEY
Approach	I-KEY
for	O
Security	O
Against	O
Multiple	O
Adversaries	O
ABSTRACT	O
In	O
adversarial	B-KEY
multiagent	I-KEY
domains	I-KEY
,	O
security	O
,	O
commonly	O
defined	O
as	O
the	O
ability	O
to	O
deal	O
with	O
intentional	O
threats	O
from	O
other	O
agents	O
,	O
is	O
a	O
critical	O
issue	O
.	O
This	O
paper	O
focuses	O
on	O
domains	O
where	O
these	O
threats	O
come	O
from	O
unknown	O
adversaries	O
.	O
These	O
domains	O
can	O
be	O
modeled	O
as	O
Bayesian	B-KEY
games	I-KEY
;	O
much	O
work	O
has	O
been	O
done	O
on	O
finding	O
equilibria	O
for	O
such	O
games	O
.	O
However	O
,	O
it	O
is	O
often	O
the	O
case	O
in	O
multiagent	O
security	O
domains	O
that	O
one	O
agent	O
can	O
commit	O
to	O
a	O
mixed	O
strategy	O
which	O
its	O
adversaries	O
observe	O
before	O
choosing	O
their	O
own	O
strategies	O
.	O
In	O
this	O
case	O
,	O
the	O
agent	O
can	O
maximize	O
reward	O
by	O
finding	O
an	O
optimal	O
strategy	O
,	O
without	O
requiring	O
equilibrium	O
.	O
Previous	O
work	O
has	O
shown	O
this	O
problem	O
of	O
optimal	O
strategy	O
selection	O
to	O
be	O
NP-hard	B-KEY
.	O
Therefore	O
,	O
we	O
present	O
a	O
heuristic	O
called	O
ASAP	O
,	O
with	O
three	O
key	O
advantages	O
to	O
address	O
the	O
problem	O
.	O
First	O
,	O
ASAP	O
searches	O
for	O
the	O
highest-reward	O
strategy	O
,	O
rather	O
than	O
a	O
Bayes-Nash	O
equilibrium	O
,	O
allowing	O
it	O
to	O
find	O
feasible	O
strategies	O
that	O
exploit	O
the	O
natural	O
first-mover	O
advantage	O
of	O
the	O
game	O
.	O
Second	O
,	O
it	O
provides	O
strategies	O
which	O
are	O
simple	O
to	O
understand	O
,	O
represent	O
,	O
and	O
implement	O
.	O
Third	O
,	O
it	O
operates	O
directly	O
on	O
the	O
compact	O
,	O
Bayesian	B-KEY
game	I-KEY
representation	O
,	O
without	O
requiring	O
conversion	O
to	O
normal	O
form	O
.	O
We	O
provide	O
an	O
efficient	O
Mixed	O
Integer	O
Linear	O
Program	O
-LRB-	O
MILP	O
-RRB-	O
implementation	O
for	O
ASAP	O
,	O
along	O
with	O
experimental	O
results	O
illustrating	O
significant	O
speedups	O
and	O
higher	O
rewards	O
over	O
other	O
approaches	O
.	O

-LRB-	O
In	O
-RRB-	O
Stability	O
Properties	O
of	O
Limit	O
Order	O
Dynamics	O
ABSTRACT	O
We	O
study	O
the	O
stability	O
properties	O
of	O
the	O
dynamics	O
of	O
the	O
standard	O
continuous	O
limit-order	O
mechanism	O
that	O
is	O
used	O
in	O
modern	B-KEY
equity	I-KEY
markets	I-KEY
.	O
We	O
ask	O
whether	O
such	O
mechanisms	O
are	O
susceptible	O
to	O
``	O
butterfly	O
effects	O
''	O
--	O
the	O
infliction	O
of	O
large	O
changes	O
on	O
common	O
measures	O
of	O
market	O
activity	O
by	O
only	O
small	O
perturbations	O
of	O
the	O
order	O
sequence	O
.	O
We	O
show	O
that	O
the	O
answer	O
depends	O
strongly	O
on	O
whether	O
the	O
market	O
consists	O
of	O
``	O
absolute	O
''	O
traders	O
-LRB-	O
who	O
determine	O
their	O
prices	O
independent	O
of	O
the	O
current	O
order	O
book	O
state	O
-RRB-	O
or	O
``	O
relative	O
''	O
traders	O
-LRB-	O
who	O
determine	O
their	O
prices	O
relative	O
to	O
the	O
current	O
bid	B-KEY
and	O
ask	O
-RRB-	O
.	O
We	O
prove	O
that	O
while	O
the	O
absolute	B-KEY
trader	I-KEY
model	I-KEY
enjoys	O
provably	O
strong	O
stability	O
properties	O
,	O
the	O
relative	B-KEY
trader	I-KEY
model	I-KEY
is	O
vulnerable	O
to	O
great	O
instability	O
.	O
Our	O
theoretical	O
results	O
are	O
supported	O
by	O
large-scale	O
experiments	O
using	O
limit	O
order	O
data	O
from	O
INET	O
,	O
a	O
large	O
electronic	O
exchange	O
for	O
NASDAQ	O
stocks	O
.	O

Congestion	B-KEY
Games	I-KEY
with	O
Load-Dependent	B-KEY
Failures	I-KEY
:	O
Identical	B-KEY
Resources	I-KEY
ABSTRACT	O
We	O
define	O
a	O
new	O
class	O
of	O
games	O
,	O
congestion	B-KEY
games	I-KEY
with	O
loaddependent	O
failures	O
-LRB-	O
CGLFs	O
-RRB-	O
,	O
which	O
generalizes	O
the	O
well-known	O
class	O
of	O
congestion	B-KEY
games	I-KEY
,	O
by	O
incorporating	O
the	O
issue	O
of	O
resource	O
failures	O
into	O
congestion	B-KEY
games	I-KEY
.	O
In	O
a	O
CGLF	O
,	O
agents	O
share	O
a	O
common	O
set	O
of	O
resources	O
,	O
where	O
each	O
resource	O
has	O
a	O
cost	O
and	O
a	O
probability	O
of	O
failure	O
.	O
Each	O
agent	O
chooses	O
a	O
subset	O
of	O
the	O
resources	O
for	O
the	O
execution	O
of	O
his	O
task	O
,	O
in	O
order	O
to	O
maximize	O
his	O
own	O
utility	O
.	O
The	O
utility	O
of	O
an	O
agent	O
is	O
the	O
difference	O
between	O
his	O
benefit	O
from	O
successful	O
task	O
completion	O
and	O
the	O
sum	O
of	O
the	O
costs	O
over	O
the	O
resources	O
he	O
uses	O
.	O
CGLFs	O
possess	O
two	O
novel	O
features	O
.	O
It	O
is	O
the	O
first	O
model	O
to	O
incorporate	O
failures	O
into	O
congestion	O
settings	O
,	O
which	O
results	O
in	O
a	O
strict	O
generalization	O
of	O
congestion	B-KEY
games	I-KEY
.	O
In	O
addition	O
,	O
it	O
is	O
the	O
first	O
model	O
to	O
consider	O
load-dependent	B-KEY
failures	I-KEY
in	O
such	O
framework	O
,	O
where	O
the	O
failure	B-KEY
probability	I-KEY
of	O
each	O
resource	O
depends	O
on	O
the	O
number	O
of	O
agents	O
selecting	O
this	O
resource	O
.	O
Although	O
,	O
as	O
we	O
show	O
,	O
CGLFs	O
do	O
not	O
admit	O
a	O
potential	B-KEY
function	I-KEY
,	O
and	O
in	O
general	O
do	O
not	O
have	O
a	O
pure	B-KEY
strategy	I-KEY
Nash	I-KEY
equilibrium	I-KEY
,	O
our	O
main	O
theorem	O
proves	O
the	O
existence	O
of	O
a	O
pure	O
strategy	O
Nash	O
equilibrium	O
in	O
every	O
CGLF	O
with	O
identical	O
resources	O
and	O
nondecreasing	O
cost	O
functions	O
.	O

An	O
Agent-Based	O
Approach	O
for	O
Privacy-Preserving	O
Recommender	B-KEY
Systems	I-KEY
ABSTRACT	O
Recommender	B-KEY
Systems	I-KEY
are	O
used	O
in	O
various	O
domains	O
to	O
generate	O
personalized	O
information	O
based	O
on	O
personal	O
user	O
data	O
.	O
The	O
ability	O
to	O
preserve	O
the	O
privacy	B-KEY
of	O
all	O
participants	O
is	O
an	O
essential	O
requirement	O
of	O
the	O
underlying	O
Information	B-KEY
Filtering	I-KEY
architectures	O
,	O
because	O
the	O
deployed	O
Recommender	B-KEY
Systems	I-KEY
have	O
to	O
be	O
accepted	O
by	O
privacy-aware	O
users	O
as	O
well	O
as	O
information	O
and	O
service	O
providers	O
.	O
Existing	O
approaches	O
neglect	O
to	O
address	O
privacy	B-KEY
in	O
this	O
multilateral	O
way	O
.	O
We	O
have	O
developed	O
an	O
approach	O
for	O
privacy-preserving	O
Recommender	B-KEY
Systems	I-KEY
based	O
on	O
Multi-Agent	O
System	O
technology	O
which	O
enables	O
applications	O
to	O
generate	O
recommendations	O
via	O
various	O
filtering	O
techniques	O
while	O
preserving	O
the	O
privacy	B-KEY
of	O
all	O
participants	O
.	O
We	O
describe	O
the	O
main	O
modules	O
of	O
our	O
solution	O
as	O
well	O
as	O
an	O
application	O
we	O
have	O
implemented	O
based	O
on	O
this	O
approach	O
.	O

DiffusionRank	B-KEY
:	O
A	O
Possible	O
Penicillin	O
for	O
Web	B-KEY
Spamming	I-KEY
ABSTRACT	O
While	O
the	O
PageRank	B-KEY
algorithm	O
has	O
proven	O
to	O
be	O
very	O
effective	O
for	O
ranking	B-KEY
Web	O
pages	O
,	O
the	O
rank	B-KEY
scores	O
of	O
Web	O
pages	O
can	O
be	O
manipulated	O
.	O
To	O
handle	O
the	O
manipulation	O
problem	O
and	O
to	O
cast	O
a	O
new	O
insight	O
on	O
the	O
Web	O
structure	O
,	O
we	O
propose	O
a	O
ranking	B-KEY
algorithm	O
called	O
DiffusionRank	B-KEY
.	O
DiffusionRank	B-KEY
is	O
motivated	O
by	O
the	O
heat	O
diffusion	O
phenomena	O
,	O
which	O
can	O
be	O
connected	O
to	O
Web	O
ranking	B-KEY
because	O
the	O
activities	O
flow	O
on	O
the	O
Web	O
can	O
be	O
imagined	O
as	O
heat	O
flow	O
,	O
the	O
link	O
from	O
a	O
page	O
to	O
another	O
can	O
be	O
treated	O
as	O
the	O
pipe	O
of	O
an	O
air-conditioner	O
,	O
and	O
heat	O
flow	O
can	O
embody	O
the	O
structure	O
of	O
the	O
underlying	O
Web	B-KEY
graph	I-KEY
.	O
Theoretically	O
we	O
show	O
that	O
DiffusionRank	B-KEY
can	O
serve	O
as	O
a	O
generalization	O
of	O
PageRank	B-KEY
when	O
the	O
heat	O
diffusion	O
coefficient	O
-	O
y	O
tends	O
to	O
infinity	O
.	O
In	O
such	O
a	O
case	O
1	O
/	O
-	O
y	O
=	O
0	O
,	O
DiffusionRank	B-KEY
-LRB-	O
PageRank	B-KEY
-RRB-	O
has	O
low	O
ability	O
of	O
anti-manipulation	O
.	O
When	O
-	O
y	O
=	O
0	O
,	O
DiffusionRank	B-KEY
obtains	O
the	O
highest	O
ability	O
of	O
anti-manipulation	O
,	O
but	O
in	O
such	O
a	O
case	O
,	O
the	O
web	O
structure	O
is	O
completely	O
ignored	O
.	O
Consequently	O
,	O
-	O
y	O
is	O
an	O
interesting	O
factor	O
that	O
can	O
control	O
the	O
balance	O
between	O
the	O
ability	O
of	O
preserving	O
the	O
original	O
Web	O
and	O
the	O
ability	O
of	O
reducing	O
the	O
effect	O
of	O
manipulation	O
.	O
It	O
is	O
found	O
empirically	O
that	O
,	O
when	O
-	O
y	O
=	O
1	O
,	O
DiffusionRank	B-KEY
has	O
a	O
Penicillin-like	O
effect	O
on	O
the	O
link	O
manipulation	O
.	O
Moreover	O
,	O
DiffusionRank	B-KEY
can	O
be	O
employed	O
to	O
find	O
group-to-group	B-KEY
relations	I-KEY
on	O
the	O
Web	O
,	O
to	O
divide	O
the	O
Web	B-KEY
graph	I-KEY
into	O
several	O
parts	O
,	O
and	O
to	O
find	O
link	B-KEY
communities	I-KEY
.	O
Experimental	O
results	O
show	O
that	O
the	O
DiffusionRank	B-KEY
algorithm	O
achieves	O
the	O
above	O
mentioned	O
advantages	O
as	O
expected	O
.	O

StarDust	O
:	O
A	O
Flexible	O
Architecture	O
for	O
Passive	O
Localization	B-KEY
in	O
Wireless	B-KEY
Sensor	I-KEY
Networks	I-KEY
*	O
Abstract	O
The	O
problem	O
of	O
localization	B-KEY
in	O
wireless	B-KEY
sensor	I-KEY
networks	I-KEY
where	O
nodes	O
do	O
not	O
use	O
ranging	B-KEY
hardware	O
,	O
remains	O
a	O
challenging	O
problem	O
,	O
when	O
considering	O
the	O
required	O
location	O
accuracy	O
,	O
energy	O
expenditure	O
and	O
the	O
duration	O
of	O
the	O
localization	B-KEY
phase	O
.	O
In	O
this	O
paper	O
we	O
propose	O
a	O
framework	O
,	O
called	O
StarDust	O
,	O
for	O
wireless	B-KEY
sensor	I-KEY
network	I-KEY
localization	B-KEY
based	O
on	O
passive	O
optical	O
components	O
.	O
In	O
the	O
StarDust	O
framework	O
,	O
sensor	B-KEY
nodes	I-KEY
are	O
equipped	O
with	O
optical	O
retro-reflectors	O
.	O
An	O
aerial	O
device	O
projects	O
light	O
towards	O
the	O
deployed	O
sensor	O
network	O
,	O
and	O
records	O
an	O
image	O
of	O
the	O
reflected	O
light	O
.	O
An	O
image	B-KEY
processing	I-KEY
algorithm	O
is	O
developed	O
for	O
obtaining	O
the	O
locations	O
of	O
sensor	B-KEY
nodes	I-KEY
.	O
For	O
matching	O
a	O
node	O
ID	O
to	O
a	O
location	O
we	O
propose	O
a	O
constraint-based	O
label	O
relaxation	O
algorithm	O
.	O
We	O
propose	O
and	O
develop	O
localization	B-KEY
techniques	O
based	O
on	O
four	O
types	O
of	O
constraints	O
:	O
node	O
color	O
,	O
neighbor	O
information	O
,	O
deployment	O
time	O
for	O
a	O
node	O
and	O
deployment	O
location	O
for	O
a	O
node	O
.	O
We	O
evaluate	O
the	O
performance	B-KEY
of	O
a	O
localization	B-KEY
system	O
based	O
on	O
our	O
framework	O
by	O
localizing	B-KEY
a	O
network	O
of	O
26	O
sensor	B-KEY
nodes	I-KEY
deployed	O
in	O
a	O
120	O
x	O
60ft2	O
area	O
.	O
The	O
localization	B-KEY
accuracy	O
ranges	B-KEY
from	O
2ft	O
to	O
5	O
ft	O
while	O
the	O
localization	B-KEY
time	O
ranges	B-KEY
from	O
10	O
milliseconds	O
to	O
2	O
minutes	O
.	O

A	O
Framework	O
for	O
Agent-Based	O
Distributed	O
Machine	O
Learning	O
and	O
Data	O
Mining	O
ABSTRACT	O
This	O
paper	O
proposes	O
a	O
framework	O
for	O
agent-based	O
distributed	O
machine	O
learning	O
and	O
data	O
mining	O
based	O
on	O
-LRB-	O
i	O
-RRB-	O
the	O
exchange	O
of	O
meta-level	O
descriptions	O
of	O
individual	O
learning	O
processes	O
among	O
agents	O
and	O
-LRB-	O
ii	O
-RRB-	O
online	O
reasoning	O
about	O
learning	O
success	O
and	O
learning	O
progress	O
by	O
learning	O
agents	O
.	O
We	O
present	O
an	O
abstract	O
architecture	O
that	O
enables	O
agents	B-KEY
to	O
exchange	O
models	O
of	O
their	O
local	O
learning	O
processes	O
and	O
introduces	O
a	O
number	O
of	O
different	O
methods	O
for	O
integrating	O
these	O
processes	O
.	O
This	O
allows	O
us	O
to	O
apply	O
existing	O
agent	B-KEY
interaction	O
mechanisms	O
to	O
distributed	B-KEY
machine	I-KEY
learning	I-KEY
tasks	O
,	O
thus	O
leveraging	O
the	O
powerful	O
coordination	O
methods	O
available	O
in	O
agent-based	O
computing	O
,	O
and	O
enables	O
agents	O
to	O
engage	O
in	O
meta-reasoning	O
about	O
their	O
own	O
learning	O
decisions	O
.	O
We	O
apply	O
this	O
architecture	O
to	O
a	O
real-world	O
distributed	B-KEY
clustering	I-KEY
application	I-KEY
to	O
illustrate	O
how	O
the	O
conceptual	O
framework	O
can	O
be	O
used	O
in	O
practical	O
systems	O
in	O
which	O
different	O
learners	O
may	O
be	O
using	O
different	O
datasets	O
,	O
hypotheses	O
and	O
learning	O
algorithms	O
.	O
We	O
report	O
on	O
experimental	O
results	O
obtained	O
using	O
this	O
system	O
,	O
review	O
related	O
work	O
on	O
the	O
subject	O
,	O
and	O
discuss	O
potential	O
future	O
extensions	O
to	O
the	O
framework	O
.	O

Hidden-Action	O
in	O
Multi-Hop	B-KEY
Routing	B-KEY
ABSTRACT	O
In	O
multi-hop	B-KEY
networks	O
,	O
the	O
actions	O
taken	O
by	O
individual	O
intermediate	O
nodes	O
are	O
typically	O
hidden	O
from	O
the	O
communicating	O
endpoints	O
;	O
all	O
the	O
endpoints	O
can	O
observe	O
is	O
whether	O
or	O
not	O
the	O
end-to-end	O
transmission	O
was	O
successful	O
.	O
Therefore	O
,	O
in	O
the	O
absence	O
of	O
incentives	B-KEY
to	O
the	O
contrary	O
,	O
rational	O
-LRB-	O
i.e.	O
,	O
selfish	O
-RRB-	O
intermediate	B-KEY
nodes	I-KEY
may	O
choose	O
to	O
forward	O
packets	O
at	O
a	O
low	O
priority	B-KEY
or	O
simply	O
not	O
forward	O
packets	O
at	O
all	O
.	O
Using	O
a	O
principal-agent	O
model	O
,	O
we	O
show	O
how	O
the	O
hidden-action	O
problem	O
can	O
be	O
overcome	O
through	O
appropriate	O
design	O
of	O
contracts	B-KEY
,	O
in	O
both	O
the	O
direct	O
-LRB-	O
the	O
endpoints	B-KEY
contract	B-KEY
with	O
each	O
individual	O
router	O
-RRB-	O
and	O
recursive	O
-LRB-	O
each	O
router	O
contracts	B-KEY
with	O
the	O
next	O
downstream	O
router	O
-RRB-	O
cases	O
.	O
We	O
further	O
demonstrate	O
that	O
per-hop	O
monitoring	O
does	O
not	O
necessarily	O
improve	O
the	O
utility	O
of	O
the	O
principal	O
or	O
the	O
social	O
welfare	O
in	O
the	O
system	O
.	O
In	O
addition	O
,	O
we	O
generalize	O
existing	O
mechanisms	B-KEY
that	O
deal	O
with	O
hidden-information	O
to	O
handle	O
scenarios	O
involving	O
both	O
hidden-information	O
and	O
hidden-action	O
.	O

Communication	B-KEY
Complexity	O
of	O
Common	O
Voting	O
Rules	O
*	O
ABSTRACT	O
We	O
determine	O
the	O
communication	B-KEY
complexity	O
of	O
the	O
common	O
voting	O
rules	O
.	O
The	O
rules	O
-LRB-	O
sorted	O
by	O
their	O
communication	B-KEY
complexity	O
from	O
low	O
to	O
high	O
-RRB-	O
are	O
plurality	O
,	O
plurality	O
with	O
runoff	O
,	O
single	O
transferable	O
vote	O
-LRB-	O
STV	O
-RRB-	O
,	O
Condorcet	O
,	O
approval	O
,	O
Bucklin	O
,	O
cup	O
,	O
maximin	O
,	O
Borda	O
,	O
Copeland	O
,	O
and	O
ranked	O
pairs	O
.	O
For	O
each	O
rule	O
,	O
we	O
first	O
give	O
a	O
deterministic	O
communication	B-KEY
protocol	B-KEY
and	O
an	O
upper	O
bound	O
on	O
the	O
number	O
of	O
bits	O
communicated	B-KEY
in	O
it	O
;	O
then	O
,	O
we	O
give	O
a	O
lower	O
bound	O
on	O
-LRB-	O
even	O
the	O
nondeterministic	O
-RRB-	O
communication	B-KEY
requirements	O
of	O
the	O
voting	B-KEY
rule	O
.	O
The	O
bounds	O
match	O
for	O
all	O
voting	B-KEY
rules	O
except	O
STV	O
and	O
maximin	O
.	O

A	O
New	O
Approach	O
for	O
Evaluating	B-KEY
Query	B-KEY
Expansion	I-KEY
:	O
Query-Document	O
Term	O
Mismatch	O
ABSTRACT	O
The	O
effectiveness	O
of	O
information	B-KEY
retrieval	I-KEY
-LRB-	O
IR	O
-RRB-	O
systems	O
is	O
influenced	O
by	O
the	O
degree	O
of	O
term	O
overlap	O
between	O
user	O
queries	O
and	O
relevant	B-KEY
documents	I-KEY
.	O
Query-document	O
term	O
mismatch	O
,	O
whether	O
partial	O
or	O
total	O
,	O
is	O
a	O
fact	O
that	O
must	O
be	O
dealt	O
with	O
by	O
IR	O
systems	O
.	O
Query	B-KEY
Expansion	I-KEY
-LRB-	O
QE	O
-RRB-	O
is	O
one	O
method	O
for	O
dealing	O
with	O
term	O
mismatch	O
.	O
IR	O
systems	O
implementing	O
query	B-KEY
expansion	I-KEY
are	O
typically	O
evaluated	B-KEY
by	O
executing	O
each	O
query	O
twice	O
,	O
with	O
and	O
without	O
query	B-KEY
expansion	I-KEY
,	O
and	O
then	O
comparing	O
the	O
two	O
result	O
sets	O
.	O
While	O
this	O
measures	O
an	O
overall	O
change	O
in	O
performance	O
,	O
it	O
does	O
not	O
directly	O
measure	O
the	O
effectiveness	O
of	O
IR	O
systems	O
in	O
overcoming	O
the	O
inherent	O
issue	O
of	O
term	O
mismatch	O
between	O
the	O
query	O
and	O
relevant	B-KEY
documents	I-KEY
,	O
nor	O
does	O
it	O
provide	O
any	O
insight	O
into	O
how	O
such	O
systems	O
would	O
behave	O
in	O
the	O
presence	O
of	O
query-document	O
term	O
mismatch	O
.	O
In	O
this	O
paper	O
,	O
we	O
propose	O
a	O
new	O
approach	O
for	O
evaluating	B-KEY
query	B-KEY
expansion	I-KEY
techniques	O
.	O
The	O
proposed	O
approach	O
is	O
attractive	O
because	O
it	O
provides	O
an	O
estimate	O
of	O
system	O
performance	O
under	O
varying	O
degrees	O
of	O
query-document	O
term	O
mismatch	O
,	O
it	O
makes	O
use	O
of	O
readily	O
available	O
test	O
collections	O
,	O
and	O
it	O
does	O
not	O
require	O
any	O
additional	O
relevance	O
judgments	O
or	O
any	O
form	O
of	O
manual	O
processing	O
.	O