A	O
High-Accuracy	O
,	O
Low-Cost	O
Localization	O
System	O
for	O
Wireless	O
Sensor	O
Networks	O
ABSTRACT	O
The	O
problem	O
of	O
localization	B-KEY
of	O
wireless	O
sensor	O
nodes	O
has	O
long	O
been	O
regarded	O
as	O
very	O
difficult	O
to	O
solve	O
,	O
when	O
considering	O
the	O
realities	O
of	O
real	O
world	O
environments	O
.	O
In	O
this	O
paper	O
,	O
we	O
formally	O
describe	O
,	O
design	O
,	O
implement	O
and	O
evaluate	O
a	O
novel	O
localization	B-KEY
system	O
,	O
called	O
Spotlight	O
.	O
Our	O
system	O
uses	O
the	O
spatio-temporal	O
properties	O
of	O
well	O
controlled	O
events	O
in	O
the	O
network	O
-LRB-	O
e.g.	O
,	O
light	O
-RRB-	O
,	O
to	O
obtain	O
the	O
locations	O
of	O
sensor	O
nodes	O
.	O
We	O
demonstrate	O
that	O
a	O
high	O
accuracy	B-KEY
in	O
localization	B-KEY
can	O
be	O
achieved	O
without	O
the	O
aid	O
of	O
expensive	O
hardware	O
on	O
the	O
sensor	O
nodes	O
,	O
as	O
required	O
by	O
other	O
localization	B-KEY
systems	O
.	O
We	O
evaluate	O
the	O
performance	B-KEY
of	O
our	O
system	O
in	O
deployments	O
of	O
Mica2	O
and	O
XSM	O
motes	O
.	O
Through	O
performance	B-KEY
evaluations	O
of	O
a	O
real	O
system	O
deployed	O
outdoors	O
,	O
we	O
obtain	O
a	O
20cm	O
localization	B-KEY
error	O
.	O
A	O
sensor	B-KEY
network	I-KEY
,	O
with	O
any	O
number	O
of	O
nodes	O
,	O
deployed	O
in	O
a	O
2500m2	O
area	O
,	O
can	O
be	O
localized	B-KEY
in	O
under	O
10	O
minutes	O
,	O
using	O
a	O
device	O
that	O
costs	O
less	O
than	O
$	O
1000	O
.	O
To	O
the	O
best	O
of	O
our	O
knowledge	O
,	O
this	O
is	O
the	O
first	O
report	O
of	O
a	O
sub-meter	O
localization	B-KEY
error	O
,	O
obtained	O
in	O
an	O
outdoor	O
environment	O
,	O
without	O
equipping	O
the	O
wireless	O
sensor	O
nodes	O
with	O
specialized	O
ranging	O
hardware	O
.	O

Distributed	O
Norm	O
Management	O
in	O
Regulated	O
Multi-Agent	O
Systems	O
*	O
ABSTRACT	O
Norms	O
are	O
widely	O
recognised	O
as	O
a	O
means	O
of	O
coordinating	B-KEY
multi-agent	O
systems	O
.	O
The	O
distributed	O
management	O
of	O
norms	O
is	O
a	O
challenging	O
issue	O
and	O
we	O
observe	O
a	O
lack	O
of	O
truly	O
distributed	O
computational	O
realisations	O
of	O
normative	O
models	O
.	O
In	O
order	O
to	O
regulate	O
the	O
behaviour	O
of	O
autonomous	O
agents	O
that	O
take	O
part	O
in	O
multiple	O
,	O
related	O
activities	B-KEY
,	O
we	O
propose	O
a	O
normative	O
model	O
,	O
the	O
Normative	B-KEY
Structure	I-KEY
-LRB-	O
NS	O
-RRB-	O
,	O
an	O
artifact	O
that	O
is	O
based	O
on	O
the	O
propagation	O
of	O
normative	B-KEY
positions	I-KEY
-LRB-	O
obligations	O
,	O
prohibitions	B-KEY
,	O
permissions	O
-RRB-	O
,	O
as	O
consequences	O
of	O
agents	O
'	O
actions	O
.	O
Within	O
a	O
NS	O
,	O
conflicts	B-KEY
may	O
arise	O
due	O
to	O
the	O
dynamic	O
nature	O
of	O
the	O
MAS	O
and	O
the	O
concurrency	O
of	O
agents	O
'	O
actions	O
.	O
However	O
,	O
ensuring	O
conflict-freedom	O
of	O
a	O
NS	O
at	O
design	O
time	O
is	O
computationally	O
intractable	O
.	O
We	O
show	O
this	O
by	O
formalising	O
the	O
notion	O
of	O
conflict	B-KEY
,	O
providing	O
a	O
mapping	O
of	O
NSs	O
into	O
Coloured	O
Petri	O
Nets	O
and	O
borrowing	O
well-known	O
theoretical	O
results	O
from	O
that	O
field	O
.	O
Since	O
online	O
conflict	B-KEY
resolution	O
is	O
required	O
,	O
we	O
present	O
a	O
tractable	O
algorithm	B-KEY
to	O
be	O
employed	O
distributedly	O
.	O
We	O
then	O
demonstrate	O
that	O
this	O
algorithm	B-KEY
is	O
paramount	O
for	O
the	O
distributed	O
enactment	O
of	O
a	O
NS	O
.	O

Interesting	B-KEY
Nuggets	O
and	O
Their	O
Impact	O
on	O
Definitional	O
Question	O
Answering	O
ABSTRACT	O
Current	O
approaches	O
to	O
identifying	O
definitional	O
sentences	O
in	O
the	O
context	O
of	O
Question	O
Answering	O
mainly	O
involve	O
the	O
use	O
of	O
linguistic	O
or	O
syntactic	O
patterns	O
to	O
identify	O
informative	B-KEY
nuggets	I-KEY
.	O
This	O
is	O
insufficient	O
as	O
they	O
do	O
not	O
address	O
the	O
novelty	O
factor	O
that	O
a	O
definitional	O
nugget	O
must	O
also	O
possess	O
.	O
This	O
paper	O
proposes	O
to	O
address	O
the	O
deficiency	O
by	O
building	O
a	O
``	O
Human	B-KEY
Interest	I-KEY
Model	O
''	O
from	O
external	O
knowledge	O
.	O
It	O
is	O
hoped	O
that	O
such	O
a	O
model	O
will	O
allow	O
the	O
computation	B-KEY
of	I-KEY
human	I-KEY
interest	I-KEY
in	O
the	O
sentence	O
with	O
respect	O
to	O
the	O
topic	O
.	O
We	O
compare	O
and	O
contrast	O
our	O
model	O
with	O
current	O
definitional	B-KEY
question	I-KEY
answering	I-KEY
models	O
to	O
show	O
that	O
interestingness	O
plays	O
an	O
important	O
factor	O
in	O
definitional	B-KEY
question	I-KEY
answering	I-KEY
.	O

Computing	O
Good	O
Nash	O
Equilibria	O
in	O
Graphical	B-KEY
Games	I-KEY
*	O
ABSTRACT	O
This	O
paper	O
addresses	O
the	O
problem	O
of	O
fair	O
equilibrium	O
selection	O
in	O
graphical	B-KEY
games	I-KEY
.	O
Our	O
approach	O
is	O
based	O
on	O
the	O
data	O
structure	O
called	O
the	O
best	O
response	O
policy	O
,	O
which	O
was	O
proposed	O
by	O
Kearns	O
et	O
al.	O
-LSB-	O
13	O
-RSB-	O
as	O
a	O
way	O
to	O
represent	O
all	O
Nash	O
equilibria	O
of	O
a	O
graphical	B-KEY
game	I-KEY
.	O
In	O
-LSB-	O
9	O
-RSB-	O
,	O
it	O
was	O
shown	O
that	O
the	O
best	O
response	O
policy	O
has	O
polynomial	O
size	O
as	O
long	O
as	O
the	O
underlying	O
graph	O
is	O
a	O
path	O
.	O
In	O
this	O
paper	O
,	O
we	O
show	O
that	O
if	O
the	O
underlying	O
graph	O
is	O
a	O
bounded-degree	O
tree	O
and	O
the	O
best	O
response	O
policy	O
has	O
polynomial	O
size	O
then	O
there	O
is	O
an	O
efficient	O
algorithm	O
which	O
constructs	O
a	O
Nash	B-KEY
equilibrium	I-KEY
that	O
guarantees	O
certain	O
payoffs	O
to	O
all	O
participants	O
.	O
Another	O
attractive	O
solution	O
concept	O
is	O
a	O
Nash	B-KEY
equilibrium	I-KEY
that	O
maximizes	O
the	O
social	B-KEY
welfare	I-KEY
.	O
We	O
show	O
that	O
,	O
while	O
exactly	O
computing	O
the	O
latter	O
is	O
infeasible	O
-LRB-	O
we	O
prove	O
that	O
solving	O
this	O
problem	O
may	O
involve	O
algebraic	O
numbers	O
of	O
an	O
arbitrarily	O
high	O
degree	O
-RRB-	O
,	O
there	O
exists	O
an	O
FPTAS	O
for	O
finding	O
such	O
an	O
equilibrium	O
as	O
long	O
as	O
the	O
best	O
response	O
policy	O
has	O
polynomial	O
size	O
.	O
These	O
two	O
algorithms	O
can	O
be	O
combined	O
to	O
produce	O
Nash	O
equilibria	O
that	O
satisfy	O
various	O
fairness	O
criteria	O
.	O

On	O
The	O
Complexity	O
of	O
Combinatorial	B-KEY
Auctions	I-KEY
:	O
Structured	B-KEY
Item	I-KEY
Graphs	I-KEY
and	O
Hypertree	B-KEY
Decompositions	I-KEY
ABSTRACT	O
The	O
winner	O
determination	O
problem	O
in	O
combinatorial	B-KEY
auctions	I-KEY
is	O
the	O
problem	O
of	O
determining	O
the	O
allocation	O
of	O
the	O
items	O
among	O
the	O
bidders	O
that	O
maximizes	O
the	O
sum	O
of	O
the	O
accepted	B-KEY
bid	I-KEY
prices	I-KEY
.	O
While	O
this	O
problem	O
is	O
in	O
general	O
NPhard	O
,	O
it	O
is	O
known	O
to	O
be	O
feasible	O
in	O
polynomial	B-KEY
time	I-KEY
on	O
those	O
instances	O
whose	O
associated	O
item	O
graphs	O
have	O
bounded	O
treewidth	O
-LRB-	O
called	O
structured	B-KEY
item	I-KEY
graphs	I-KEY
-RRB-	O
.	O
Formally	O
,	O
an	O
item	O
graph	O
is	O
a	O
graph	O
whose	O
nodes	O
are	O
in	O
one-to-one	O
correspondence	O
with	O
items	O
,	O
and	O
edges	O
are	O
such	O
that	O
for	O
any	O
bid	O
,	O
the	O
items	O
occurring	O
in	O
it	O
induce	O
a	O
connected	O
subgraph	O
.	O
Note	O
that	O
many	O
item	O
graphs	O
might	O
be	O
associated	O
with	O
a	O
given	O
combinatorial	B-KEY
auction	I-KEY
,	O
depending	O
on	O
the	O
edges	O
selected	O
for	O
guaranteeing	O
the	O
connectedness	O
.	O
In	O
fact	O
,	O
the	O
tractability	O
of	O
determining	O
whether	O
a	O
structured	B-KEY
item	I-KEY
graph	I-KEY
of	O
a	O
fixed	B-KEY
treewidth	I-KEY
exists	O
-LRB-	O
and	O
if	O
so	O
,	O
computing	O
one	O
-RRB-	O
was	O
left	O
as	O
a	O
crucial	O
open	O
problem	O
.	O
In	O
this	O
paper	O
,	O
we	O
solve	O
this	O
problem	O
by	O
proving	O
that	O
the	O
existence	O
of	O
a	O
structured	B-KEY
item	I-KEY
graph	I-KEY
is	O
computationally	O
intractable	O
,	O
even	O
for	O
treewidth	O
3	O
.	O
Motivated	O
by	O
this	O
bad	O
news	O
,	O
we	O
investigate	O
different	O
kinds	O
of	O
structural	O
requirements	O
that	O
can	O
be	O
used	O
to	O
isolate	O
tractable	O
classes	O
of	O
combinatorial	B-KEY
auctions	I-KEY
.	O
We	O
show	O
that	O
the	O
notion	O
of	O
hypertree	B-KEY
decomposition	I-KEY
,	O
a	O
recently	O
introduced	O
measure	O
of	O
hypergraph	B-KEY
cyclicity	O
,	O
turns	O
out	O
to	O
be	O
most	O
useful	O
here	O
.	O
Indeed	O
,	O
we	O
show	O
that	O
the	O
winner	O
determination	O
problem	O
is	O
solvable	O
in	O
polynomial	B-KEY
time	I-KEY
on	O
instances	O
whose	O
bidder	O
interactions	O
can	O
be	O
represented	O
with	O
-LRB-	O
dual	O
-RRB-	O
hypergraphs	B-KEY
having	O
bounded	O
hypertree	O
width	O
.	O
Even	O
more	O
surprisingly	O
,	O
we	O
show	O
that	O
the	O
class	O
of	O
tractable	O
instances	O
identified	O
by	O
means	O
of	O
our	O
approach	O
properly	O
contains	O
the	O
class	O
of	O
instances	O
having	O
a	O
structured	B-KEY
item	I-KEY
graph	I-KEY
.	O

Generalized	B-KEY
Trade	I-KEY
Reduction	I-KEY
Mechanisms	O
ABSTRACT	O
When	O
designing	O
a	O
mechanism	O
there	O
are	O
several	O
desirable	O
properties	O
to	O
maintain	O
such	O
as	O
incentive	O
compatibility	O
-LRB-	O
IC	O
-RRB-	O
,	O
individual	O
rationality	O
-LRB-	O
IR	O
-RRB-	O
,	O
and	O
budget	B-KEY
balance	I-KEY
-LRB-	O
BB	O
-RRB-	O
.	O
It	O
is	O
well	O
known	O
-LSB-	O
15	O
-RSB-	O
that	O
it	O
is	O
impossible	O
for	O
a	O
mechanism	O
to	O
maximize	O
social	O
welfare	O
whilst	O
also	O
being	O
IR	O
,	O
IC	O
,	O
and	O
BB	O
.	O
There	O
have	O
been	O
several	O
attempts	O
to	O
circumvent	O
-LSB-	O
15	O
-RSB-	O
by	O
trading	O
welfare	O
for	O
BB	O
,	O
e.g.	O
,	O
in	O
domains	O
such	O
as	O
double-sided	O
auctions	O
-LSB-	O
13	O
-RSB-	O
,	O
distributed	O
markets	O
-LSB-	O
3	O
-RSB-	O
and	O
supply	O
chain	O
problems	O
-LSB-	O
2	O
,	O
4	O
-RSB-	O
.	O
In	O
this	O
paper	O
we	O
provide	O
a	O
procedure	O
called	O
a	O
Generalized	B-KEY
Trade	I-KEY
Reduction	I-KEY
-LRB-	O
GTR	O
-RRB-	O
for	O
single-value	O
players	O
,	O
which	O
given	O
an	O
IR	O
and	O
IC	O
mechanism	O
,	O
outputs	O
a	O
mechanism	O
which	O
is	O
IR	O
,	O
IC	O
and	O
BB	O
with	O
a	O
loss	O
of	O
welfare	O
.	O
We	O
bound	O
the	O
welfare	O
achieved	O
by	O
our	O
procedure	O
for	O
a	O
wide	O
range	O
of	O
domains	O
.	O
In	O
particular	O
,	O
our	O
results	O
improve	O
on	O
existing	O
solutions	O
for	O
problems	O
such	O
as	O
double	O
sided	O
markets	O
with	O
homogenous	B-KEY
goods	I-KEY
,	O
distributed	O
markets	O
and	O
several	O
kinds	O
of	O
supply	O
chains	O
.	O
Furthermore	O
,	O
our	O
solution	O
provides	O
budget	B-KEY
balanced	I-KEY
mechanisms	O
for	O
several	O
open	O
problems	O
such	O
as	O
combinatorial	O
double-sided	O
auctions	O
and	O
distributed	O
markets	O
with	O
strategic	O
transportation	O
edges	O
.	O

Fast	O
Generation	O
of	O
Result	O
Snippets	O
in	O
Web	O
Search	O
ABSTRACT	O
The	O
presentation	O
of	O
query	O
biased	O
document	O
snippets	O
as	O
part	O
of	O
results	O
pages	O
presented	O
by	O
search	B-KEY
engines	I-KEY
has	O
become	O
an	O
expectation	O
of	O
search	B-KEY
engine	I-KEY
users	O
.	O
In	O
this	O
paper	O
we	O
explore	O
the	O
algorithms	O
and	O
data	O
structures	O
required	O
as	O
part	O
of	O
a	O
search	B-KEY
engine	I-KEY
to	O
allow	O
efficient	O
generation	O
of	O
query	O
biased	O
snippets	O
.	O
We	O
begin	O
by	O
proposing	O
and	O
analysing	O
a	O
document	O
compression	O
method	O
that	O
reduces	O
snippet	B-KEY
generation	I-KEY
time	O
by	O
58	O
%	O
over	O
a	O
baseline	O
using	O
the	O
zlib	O
compression	O
library	O
.	O
These	O
experiments	O
reveal	O
that	O
finding	O
documents	O
on	O
secondary	O
storage	O
dominates	O
the	O
total	O
cost	O
of	O
generating	O
snippets	O
,	O
and	O
so	O
caching	O
documents	O
in	O
RAM	B-KEY
is	O
essential	O
for	O
a	O
fast	O
snippet	B-KEY
generation	I-KEY
process	O
.	O
Using	O
simulation	O
,	O
we	O
examine	O
snippet	B-KEY
generation	I-KEY
performance	B-KEY
for	O
different	O
size	O
RAM	B-KEY
caches	O
.	O
Finally	O
we	O
propose	O
and	O
analyse	O
document	O
reordering	O
and	O
compaction	O
,	O
revealing	O
a	O
scheme	O
that	O
increases	O
the	O
number	O
of	O
document	B-KEY
cache	I-KEY
hits	O
with	O
only	O
a	O
marginal	O
affect	O
on	O
snippet	O
quality	O
.	O
This	O
scheme	O
effectively	O
doubles	O
the	O
number	O
of	O
documents	O
that	O
can	O
fit	O
in	O
a	O
fixed	O
size	O
cache	O
.	O

Deployment	O
Issues	O
of	O
a	O
VoIP	B-KEY
Conferencing	O
System	O
in	O
a	O
Virtual	O
Conferencing	O
Environment	O
ABSTRACT	O
Real-time	O
services	O
have	O
been	O
supported	O
by	O
and	O
large	O
on	O
circuitswitched	O
networks	O
.	O
Recent	O
trends	O
favour	O
services	O
ported	O
on	O
packet-switched	B-KEY
networks	I-KEY
.	O
For	O
audio	O
conferencing	O
,	O
we	O
need	O
to	O
consider	O
many	O
issues	O
--	O
scalability	O
,	O
quality	O
of	O
the	O
conference	O
application	O
,	O
floor	O
control	O
and	O
load	O
on	O
the	O
clients/servers	O
--	O
to	O
name	O
a	O
few	O
.	O
In	O
this	O
paper	O
,	O
we	O
describe	O
an	O
audio	B-KEY
service	I-KEY
framework	I-KEY
designed	O
to	O
provide	O
a	O
Virtual	B-KEY
Conferencing	I-KEY
Environment	I-KEY
-LRB-	O
VCE	B-KEY
-RRB-	O
.	O
The	O
system	O
is	O
designed	O
to	O
accommodate	O
a	O
large	O
number	O
of	O
end	O
users	O
speaking	O
at	O
the	O
same	O
time	O
and	O
spread	O
across	O
the	O
Internet	O
.	O
The	O
framework	O
is	O
based	O
on	O
Conference	B-KEY
Servers	I-KEY
-LSB-	O
14	O
-RSB-	O
,	O
which	O
facilitate	O
the	O
audio	O
handling	O
,	O
while	O
we	O
exploit	O
the	O
SIP	B-KEY
capabilities	O
for	O
signaling	O
purposes	O
.	O
Client	O
selection	O
is	O
based	O
on	O
a	O
recent	O
quantifier	O
called	O
``	O
Loudness	B-KEY
Number	I-KEY
''	O
that	O
helps	O
mimic	O
a	O
physical	O
face-to-face	O
conference	O
.	O
We	O
deal	O
with	O
deployment	O
issues	O
of	O
the	O
proposed	O
solution	O
both	O
in	O
terms	O
of	O
scalability	O
and	O
interactivity	O
,	O
while	O
explaining	O
the	O
techniques	O
we	O
use	O
to	O
reduce	O
the	O
traffic	O
.	O
We	O
have	O
implemented	O
a	O
Conference	B-KEY
Server	I-KEY
-LRB-	O
CS	O
-RRB-	O
application	O
on	O
a	O
campus-wide	O
network	O
at	O
our	O
Institute	O
.	O

Intra-flow	O
Loss	B-KEY
Recovery	I-KEY
and	I-KEY
Control	I-KEY
for	O
ABSTRACT	O
``	O
Best	O
effort	O
''	O
packet-switched	O
networks	O
,	O
like	O
the	O
Internet	O
,	O
do	O
not	O
offer	O
a	O
reliable	O
transmission	O
of	O
packets	O
to	O
applications	O
with	O
real-time	O
constraints	O
such	O
voice	O
.	O
Thus	O
,	O
the	O
loss	O
of	O
packets	O
impairs	O
the	O
application-level	O
utility	O
.	O
For	O
voice	O
this	O
utility	O
impairment	O
is	O
twofold	O
:	O
on	O
one	O
hand	O
,	O
even	O
short	O
bursts	O
of	O
lost	O
packets	O
may	O
decrease	O
significantly	O
the	O
ability	O
of	O
the	O
receiver	O
to	O
conceal	O
the	O
packet	O
loss	O
and	O
the	O
speech	O
signal	O
out	O
is	O
interrupted	O
.	O
On	O
the	O
other	O
hand	O
,	O
some	O
packets	O
may	O
be	O
particular	O
sensitive	O
to	O
loss	O
as	O
they	O
carry	O
more	O
important	O
information	O
in	O
terms	O
of	O
user	O
perception	O
than	O
other	O
packets	O
.	O
We	O
first	O
develop	O
an	O
end-to-end	B-KEY
model	I-KEY
based	O
on	O
loss	O
lengths	O
with	O
which	O
we	O
can	O
describe	O
the	O
loss	O
distribution	O
within	O
a	O
These	O
packet-level	B-KEY
metrics	I-KEY
are	O
then	O
linked	O
to	O
user-level	O
objective	O
speech	O
quality	O
metrics	O
.	O
Using	O
this	O
framework	O
,	O
we	O
find	O
that	O
for	O
low-compressing	O
sample-based	O
codecs	O
-LRB-	O
PCM	O
-RRB-	O
with	O
loss	B-KEY
concealment	I-KEY
isolated	O
packet	O
losses	O
can	O
be	O
concealed	O
well	O
,	O
whereas	O
burst	O
losses	O
have	O
a	O
higher	O
perceptual	O
impact	O
.	O
For	O
high-compressing	O
frame-based	O
codecs	O
-LRB-	O
G.	O
729	O
-RRB-	O
on	O
one	O
hand	O
the	O
impact	O
of	O
loss	O
is	O
amplified	O
through	O
error	O
propagation	O
caused	O
by	O
the	O
decoder	O
filter	O
memories	O
,	O
though	O
on	O
the	O
other	O
hand	O
such	O
coding	O
schemes	O
help	O
to	O
perform	O
loss	B-KEY
concealment	I-KEY
by	O
extrapolation	O
of	O
decoder	O
state	O
.	O
Contrary	O
to	O
sample-based	O
codecs	O
we	O
show	O
that	O
the	O
concealment	O
performance	O
may	O
``	O
break	O
''	O
at	O
transitions	O
within	O
the	O
speech	O
signal	O
however	O
.	O
We	O
then	O
propose	O
mechanisms	O
which	O
differentiate	O
between	O
packets	O
within	O
a	O
voice	O
data	O
to	O
minimize	O
the	O
impact	O
of	O
packet	O
loss	O
.	O
We	O
designate	O
these	O
methods	O
as	O
loss	B-KEY
recovery	I-KEY
and	I-KEY
control	I-KEY
.	O
At	O
the	O
end-to-end	O
level	O
,	O
identification	O
of	O
packets	O
sensitive	O
to	O
loss	O
-LRB-	O
sender	O
-RRB-	O
as	O
well	O
as	O
loss	B-KEY
concealment	I-KEY
-LRB-	O
receiver	O
-RRB-	O
takes	O
place	O
.	O
Hop-by-hop	O
support	O
schemes	O
then	O
allow	O
to	O
-LRB-	O
statistically	O
-RRB-	O
trade	O
the	O
loss	O
of	O
one	O
packet	O
,	O
which	O
is	O
considered	O
more	O
important	O
,	O
against	O
another	O
one	O
of	O
the	O
same	O
flow	O
which	O
is	O
of	O
lower	O
importance	O
.	O
As	O
both	O
ets	O
require	O
the	O
same	O
cost	O
in	O
terms	O
of	O
network	O
transmission	O
,	O
a	O
gain	O
in	O
user	O
perception	O
is	O
obtainable	O
.	O
We	O
show	O
that	O
significant	O
speech	O
quality	O
improvements	O
can	O
be	O
achieved	O
and	O
additional	O
data	O
and	O
delay	O
overhead	O
can	O
be	O
avoided	O
while	O
still	O
maintaining	O
a	O
network	O
service	O
which	O
is	O
virtually	O
identical	O
to	O
best	O
effort	O
in	O
the	O
long	O
term	O
.	O

Estimation	O
and	O
Use	O
of	O
Uncertainty	O
in	O
Pseudo-relevance	B-KEY
Feedback	I-KEY
ABSTRACT	O
Existing	O
pseudo-relevance	B-KEY
feedback	I-KEY
methods	O
typically	O
perform	O
averaging	O
over	O
the	O
top-retrieved	O
documents	O
,	O
but	O
ignore	O
an	O
important	O
statistical	O
dimension	O
:	O
the	O
risk	O
or	O
variance	O
associated	O
with	O
either	O
the	O
individual	O
document	O
models	O
,	O
or	O
their	O
combination	O
.	O
Treating	O
the	O
baseline	O
feedback	B-KEY
method	I-KEY
as	O
a	O
black	O
box	O
,	O
and	O
the	O
output	O
feedback	B-KEY
model	I-KEY
as	O
a	O
random	O
variable	O
,	O
we	O
estimate	O
a	O
posterior	B-KEY
distribution	I-KEY
for	O
the	O
feedback	B-KEY
model	I-KEY
by	O
resampling	O
a	O
given	O
query	O
's	O
top-retrieved	O
documents	O
,	O
using	O
the	O
posterior	O
mean	O
or	O
mode	O
as	O
the	O
enhanced	B-KEY
feedback	I-KEY
model	I-KEY
.	O
We	O
then	O
perform	O
model	O
combination	O
over	O
several	O
enhanced	O
models	O
,	O
each	O
based	O
on	O
a	O
slightly	O
modified	O
query	O
sampled	O
from	O
the	O
original	O
query	O
.	O
We	O
find	O
that	O
resampling	O
documents	O
helps	O
increase	O
individual	O
feedback	B-KEY
model	I-KEY
precision	O
by	O
removing	O
noise	O
terms	O
,	O
while	O
sampling	O
from	O
the	O
query	O
improves	O
robustness	O
-LRB-	O
worst-case	O
performance	O
-RRB-	O
by	O
emphasizing	O
terms	O
related	O
to	O
multiple	O
query	O
aspects	O
.	O
The	O
result	O
is	O
a	O
meta-feedback	O
algorithm	O
that	O
is	O
both	O
more	O
robust	O
and	O
more	O
precise	O
than	O
the	O
original	O
strong	O
baseline	O
method	O
.	O

Pruning	B-KEY
Policies	O
for	O
Two-Tiered	O
Inverted	B-KEY
Index	I-KEY
with	O
Correctness	B-KEY
Guarantee	I-KEY
ABSTRACT	O
The	O
Web	B-KEY
search	I-KEY
engines	I-KEY
maintain	O
large-scale	O
inverted	B-KEY
indexes	I-KEY
which	O
are	O
queried	O
thousands	O
of	O
times	O
per	O
second	O
by	O
users	O
eager	O
for	O
information	O
.	O
In	O
order	O
to	O
cope	O
with	O
the	O
vast	O
amounts	O
of	O
query	B-KEY
loads	I-KEY
,	O
search	O
engines	O
prune	B-KEY
their	O
index	O
to	O
keep	O
documents	O
that	O
are	O
likely	O
to	O
be	O
returned	O
as	O
top	O
results	O
,	O
and	O
use	O
this	O
pruned	B-KEY
index	O
to	O
compute	O
the	O
first	O
batches	O
of	O
results	O
.	O
While	O
this	O
approach	O
can	O
improve	O
performance	O
by	O
reducing	O
the	O
size	O
of	O
the	O
index	O
,	O
if	O
we	O
compute	O
the	O
top	O
results	O
only	O
from	O
the	O
pruned	B-KEY
index	O
we	O
may	O
notice	O
a	O
significant	O
degradation	O
in	O
the	O
result	O
quality	O
:	O
if	O
a	O
document	O
should	O
be	O
in	O
the	O
top	O
results	O
but	O
was	O
not	O
included	O
in	O
the	O
pruned	O
index	O
,	O
it	O
will	O
be	O
placed	O
behind	O
the	O
results	O
computed	O
from	O
the	O
pruned	O
index	O
.	O
Given	O
the	O
fierce	O
competition	O
in	O
the	O
online	B-KEY
search	I-KEY
market	I-KEY
,	O
this	O
phenomenon	O
is	O
clearly	O
undesirable	O
.	O
In	O
this	O
paper	O
,	O
we	O
study	O
how	O
we	O
can	O
avoid	O
any	O
degradation	B-KEY
of	I-KEY
result	I-KEY
quality	I-KEY
due	O
to	O
the	O
pruning-based	O
performance	O
optimization	O
,	O
while	O
still	O
realizing	O
most	O
of	O
its	O
benefit	O
.	O
Our	O
contribution	O
is	O
a	O
number	O
of	O
modifications	O
in	O
the	O
pruning	B-KEY
techniques	O
for	O
creating	O
the	O
pruned	O
index	O
and	O
a	O
new	O
result	O
computation	O
algorithm	O
that	O
guarantees	O
that	O
the	O
top-matching	O
pages	O
are	O
always	O
placed	O
at	O
the	O
top	O
search	O
results	O
,	O
even	O
though	O
we	O
are	O
computing	O
the	O
first	O
batch	O
from	O
the	O
pruned	O
index	O
most	O
of	O
the	O
time	O
.	O
We	O
also	O
show	O
how	O
to	O
determine	O
the	O
optimal	B-KEY
size	I-KEY
of	O
a	O
pruned	B-KEY
index	O
and	O
we	O
experimentally	O
evaluate	O
our	O
algorithms	O
on	O
a	O
collection	O
of	O
130	O
million	O
Web	O
pages	O
.	O

Budget	B-KEY
Optimization	I-KEY
in	O
Search-Based	O
Advertising	O
Auctions	O
ABSTRACT	O
Internet	B-KEY
search	O
companies	O
sell	O
advertisement	B-KEY
slots	O
based	O
on	O
users	O
'	O
search	O
queries	O
via	O
an	O
auction	B-KEY
.	O
While	O
there	O
has	O
been	O
previous	O
work	O
on	O
the	O
auction	B-KEY
process	O
and	O
its	O
game-theoretic	O
aspects	O
,	O
most	O
of	O
it	O
focuses	O
on	O
the	O
Internet	B-KEY
company	O
.	O
In	O
this	O
work	O
,	O
we	O
focus	O
on	O
the	O
advertisers	B-KEY
,	O
who	O
must	O
solve	O
a	O
complex	O
optimization	B-KEY
problem	O
to	O
decide	O
how	O
to	O
place	O
bids	B-KEY
on	O
keywords	B-KEY
to	O
maximize	O
their	O
return	O
-LRB-	O
the	O
number	O
of	O
user	O
clicks	O
on	O
their	O
ads	O
-RRB-	O
for	O
a	O
given	O
budget	O
.	O
We	O
model	O
the	O
entire	O
process	O
and	O
study	O
this	O
budget	B-KEY
optimization	I-KEY
problem	O
.	O
While	O
most	O
variants	O
are	O
NP-hard	O
,	O
we	O
show	O
,	O
perhaps	O
surprisingly	O
,	O
that	O
simply	O
randomizing	O
between	O
two	O
uniform	O
strategies	O
that	O
bid	B-KEY
equally	O
on	O
all	O
the	O
keywords	B-KEY
works	O
well	O
.	O
More	O
precisely	O
,	O
this	O
strategy	O
gets	O
at	O
least	O
a	O
1	O
−	O
1/e	O
fraction	O
of	O
the	O
maximum	O
clicks	O
possible	O
.	O
As	O
our	O
preliminary	O
experiments	O
show	O
,	O
such	O
uniform	O
strategies	O
are	O
likely	O
to	O
be	O
practical	O
.	O
We	O
also	O
present	O
inapproximability	O
results	O
,	O
and	O
optimal	B-KEY
algorithms	O
for	O
variants	O
of	O
the	O
budget	B-KEY
optimization	I-KEY
problem	O
.	O

Term	O
Feedback	O
for	O
Information	B-KEY
Retrieval	I-KEY
with	O
Language	B-KEY
Models	I-KEY
ABSTRACT	O
I	O
n	O
t	O
hi	O
s	O
paper	O
w	O
e	O
s	O
t	O
udy	O
t	O
er	O
m	O
-	O
based	O
f	O
eedback	O
f	O
or	O
i	O
nf	O
or	O
mat	O
i	O
on	O
r	O
etrieval	O
in	O
the	O
language	B-KEY
modeling	I-KEY
approach	O
.	O
With	O
term	O
feedback	O
a	O
user	O
directly	O
judges	O
the	O
relevance	O
of	O
individual	O
terms	O
without	O
interaction	O
with	O
feedback	O
documents	O
,	O
taking	O
full	O
control	O
of	O
the	O
query	B-KEY
expansion	I-KEY
process	O
.	O
We	O
propose	O
a	O
cluster-based	O
method	O
for	O
selecting	O
terms	O
to	O
present	O
to	O
the	O
user	O
for	O
judgment	O
,	O
as	O
well	O
as	O
effective	O
algorithms	O
for	O
constructing	O
refined	O
query	O
language	B-KEY
models	I-KEY
from	O
user	O
term	O
feedback	O
.	O
Our	O
algorithms	O
are	O
shown	O
to	O
bring	O
significant	O
improvement	O
in	O
retrieval	O
accuracy	O
over	O
a	O
non-feedback	O
baseline	O
,	O
and	O
achieve	O
comparable	O
performance	O
to	O
relevance	O
feedback	O
.	O
They	O
are	O
helpful	O
even	O
when	O
there	O
are	O
no	O
relevant	O
documents	O
in	O
the	O
top	O
.	O

Towards	O
Self-organising	B-KEY
Agent-based	O
Resource	O
Allocation	O
in	O
a	O
Multi-Server	O
Environment	O
ABSTRACT	O
Distributed	O
applications	O
require	O
distributed	O
techniques	O
for	O
efficient	O
resource	B-KEY
allocation	I-KEY
.	O
These	O
techniques	O
need	O
to	O
take	O
into	O
account	O
the	O
heterogeneity	O
and	O
potential	O
unreliability	O
of	O
resources	O
and	O
resource	O
consumers	O
in	O
a	O
distributed	O
environments	O
.	O
In	O
this	O
paper	O
we	O
propose	O
a	O
distributed	B-KEY
algorithm	I-KEY
that	O
solves	O
the	O
resource	B-KEY
allocation	I-KEY
problem	O
in	O
distributed	O
multiagent	O
systems	O
.	O
Our	O
solution	O
is	O
based	O
on	O
the	O
self-organisation	B-KEY
of	O
agents	B-KEY
,	O
which	O
does	O
not	O
require	O
any	O
facilitator	O
or	O
management	O
layer	O
.	O
The	O
resource	B-KEY
allocation	I-KEY
in	O
the	O
system	O
is	O
a	O
purely	O
emergent	O
effect	O
.	O
We	O
present	O
results	O
of	O
the	O
proposed	O
resource	B-KEY
allocation	I-KEY
mechanism	O
in	O
the	O
simulated	O
static	O
and	O
dynamic	O
multi-server	O
environment	O
.	O

Realistic	O
Cognitive	O
Load	O
Modeling	O
for	O
Enhancing	O
Shared	O
Mental	O
Models	O
in	O
Human-Agent	O
Collaboration	B-KEY
ABSTRACT	O
Human	O
team	O
members	O
often	O
develop	O
shared	O
expectations	B-KEY
to	O
predict	O
each	O
other	O
's	O
needs	O
and	O
coordinate	O
their	O
behaviors	O
.	O
In	O
this	O
paper	O
the	O
concept	O
``	O
Shared	B-KEY
Belief	I-KEY
Map	I-KEY
''	O
is	O
proposed	O
as	O
a	O
basis	O
for	O
developing	O
realistic	O
shared	O
expectations	B-KEY
among	O
a	O
team	O
of	O
Human-Agent-Pairs	O
-LRB-	O
HAPs	O
-RRB-	O
.	O
The	O
establishment	O
of	O
shared	B-KEY
belief	I-KEY
maps	I-KEY
relies	O
on	O
inter-agent	O
information	O
sharing	O
,	O
the	O
effectiveness	O
of	O
which	O
highly	O
depends	O
on	O
agents	O
'	O
processing	O
loads	O
and	O
the	O
instantaneous	O
cognitive	O
loads	O
of	O
their	O
human	O
partners	O
.	O
We	O
investigate	O
HMM-based	O
cognitive	O
load	O
models	O
to	O
facilitate	O
team	O
members	O
to	O
``	O
share	O
the	O
right	O
information	O
with	O
the	O
right	O
party	O
at	O
the	O
right	O
time	O
''	O
.	O
The	O
shared	B-KEY
belief	I-KEY
map	I-KEY
concept	O
and	O
the	O
cognitive/processing	O
load	O
models	O
have	O
been	O
implemented	O
in	O
a	O
cognitive	O
agent	O
architecture	O
--	O
SMMall	O
.	O
A	O
series	O
of	O
experiments	O
were	O
conducted	O
to	O
evaluate	O
the	O
concept	O
,	O
the	O
models	O
,	O
and	O
their	O
impacts	O
on	O
the	O
evolving	O
of	O
shared	O
mental	O
models	O
of	O
HAP	O
teams	O
.	O

Service	B-KEY
Interface	I-KEY
:	O
A	O
New	O
Abstraction	O
for	O
Implementing	O
and	O
Composing	O
Protocols	O
*	O
ABSTRACT	O
In	O
this	O
paper	O
we	O
compare	O
two	O
approaches	O
to	O
the	O
design	O
of	O
protocol	B-KEY
frameworks	I-KEY
--	O
tools	O
for	O
implementing	O
modular	B-KEY
network	B-KEY
protocols	O
.	O
The	O
most	O
common	O
approach	O
uses	O
events	O
as	O
the	O
main	O
abstraction	O
for	O
a	O
local	O
interaction	O
between	O
protocol	O
modules	B-KEY
.	O
We	O
argue	O
that	O
an	O
alternative	O
approach	O
,	O
that	O
is	O
based	O
on	O
service	O
abstraction	O
,	O
is	O
more	O
suitable	O
for	O
expressing	O
modular	B-KEY
protocols	O
.	O
It	O
also	O
facilitates	O
advanced	O
features	O
in	O
the	O
design	O
of	O
protocols	O
,	O
such	O
as	O
dynamic	O
update	O
of	O
distributed	O
protocols	O
.	O
We	O
then	O
describe	O
an	O
experimental	O
implementation	O
of	O
a	O
service-based	O
protocol	B-KEY
framework	I-KEY
in	O
Java	O
.	O

A	O
Formal	O
Road	O
from	O
Institutional	B-KEY
Norms	O
to	O
Organizational	O
Structures	O
ABSTRACT	O
Up	O
to	O
now	O
,	O
the	O
way	O
institutions	B-KEY
and	O
organizations	O
have	O
been	O
used	O
in	O
the	O
development	O
of	O
open	O
systems	O
has	O
not	O
often	O
gone	O
further	O
than	O
a	O
useful	O
heuristics	O
.	O
In	O
order	O
to	O
develop	O
systems	O
actually	O
implementing	O
institutions	B-KEY
and	O
organizations	O
,	O
formal	B-KEY
methods	I-KEY
should	O
take	O
the	O
place	O
of	O
heuristic	O
ones	O
.	O
The	O
paper	O
presents	O
a	O
formal	O
semantics	O
for	O
the	O
notion	O
of	O
institution	B-KEY
and	O
its	O
components	O
-LRB-	O
abstract	O
and	O
concrete	O
norms	B-KEY
,	O
empowerment	O
of	O
agents	O
,	O
roles	B-KEY
-RRB-	O
and	O
defines	O
a	O
formal	O
relation	O
between	O
institutions	B-KEY
and	O
organizational	B-KEY
structures	I-KEY
.	O
As	O
a	O
result	O
,	O
it	O
is	O
shown	O
how	O
institutional	B-KEY
norms	O
can	O
be	O
refined	O
to	O
constructs	O
--	O
organizational	O
structures	O
--	O
which	O
are	O
closer	O
to	O
an	O
implemented	O
system	O
.	O
It	O
is	O
also	O
shown	O
how	O
such	O
a	O
refinement	O
process	O
can	O
be	O
fully	O
formalized	O
and	O
it	O
is	O
therefore	O
amenable	O
to	O
rigorous	O
verification	O
.	O

Approximately-Strategyproof	O
and	O
Tractable	O
Multi-Unit	O
Auctions	O
ABSTRACT	O
We	O
present	O
an	O
approximately-efficient	O
and	O
approximatelystrategyproof	O
auction	O
mechanism	O
for	O
a	O
single-good	O
multi-unit	O
allocation	O
problem	O
.	O
The	O
bidding	B-KEY
language	I-KEY
in	O
our	O
auctions	O
allows	O
marginal-decreasing	O
piecewise	O
constant	O
curves	O
.	O
First	O
,	O
we	O
develop	O
a	O
fully	O
polynomial-time	O
approximation	O
scheme	O
for	O
the	O
multi-unit	O
allocation	O
problem	O
,	O
which	O
computes	O
a	O
-LRB-	O
1	O
+	O
e	O
-RRB-	O
approximation	O
in	O
worst-case	O
time	O
T	O
=	O
O	O
-LRB-	O
n3/e	O
-RRB-	O
,	O
given	O
n	O
bids	O
each	O
with	O
a	O
constant	O
number	O
of	O
pieces	O
.	O
Second	O
,	O
we	O
embed	O
this	O
approximation	O
scheme	O
within	O
a	O
Vickrey-Clarke-Groves	O
-LRB-	O
VCG	O
-RRB-	O
mechanism	O
and	O
compute	O
payments	O
to	O
n	O
agents	O
for	O
an	O
asymptotic	O
cost	O
of	O
O	O
-LRB-	O
T	O
log	O
n	O
-RRB-	O
.	O
The	O
maximal	O
possible	O
gain	O
from	O
manipulation	O
to	O
a	O
bidder	O
in	O
the	O
combined	O
scheme	O
is	O
bounded	O
by	O
e	O
/	O
-LRB-	O
1	O
+	O
e	O
-RRB-	O
V	O
,	O
where	O
V	O
is	O
the	O
total	O
surplus	O
in	O
the	O
efficient	O
outcome	O
.	O

BuddyCache	B-KEY
:	O
High-Performance	O
Object	O
Storage	O
for	O
Collaborative	B-KEY
Strong-Consistency	I-KEY
Applications	I-KEY
in	O
a	O
WAN	O
*	O
ABSTRACT	O
Collaborative	O
applications	O
provide	O
a	O
shared	O
work	O
environment	O
for	O
groups	O
of	O
networked	O
clients	O
collaborating	O
on	O
a	O
common	O
task	O
.	O
They	O
require	O
strong	O
consistency	O
for	O
shared	O
persistent	O
data	O
and	O
efficient	O
access	O
to	O
fine-grained	O
objects	O
.	O
These	O
properties	O
are	O
difficult	O
to	O
provide	O
in	O
wide-area	B-KEY
networks	I-KEY
because	O
of	O
high	O
network	O
latency	O
.	O
BuddyCache	B-KEY
is	O
a	O
new	O
transactional	B-KEY
caching	O
approach	O
that	O
improves	O
the	O
latency	O
of	O
access	O
to	O
shared	O
persistent	O
objects	O
for	O
collaborative	B-KEY
strong-consistency	I-KEY
applications	I-KEY
in	O
high-latency	O
network	O
environments	O
.	O
The	O
challenge	O
is	O
to	O
improve	O
performance	O
while	O
providing	O
the	O
correctness	O
and	O
availability	O
properties	O
of	O
a	O
transactional	B-KEY
caching	O
protocol	O
in	O
the	O
presence	O
of	O
node	O
failures	O
and	O
slow	O
peers	O
.	O
We	O
have	O
implemented	O
a	O
BuddyCache	B-KEY
prototype	O
and	O
evaluated	O
its	O
performance	O
.	O
Analytical	O
results	O
,	O
confirmed	O
by	O
measurements	O
of	O
the	O
BuddyCache	B-KEY
prototype	O
using	O
the	O
multiuser	O
007	O
benchmark	O
indicate	O
that	O
for	O
typical	O
Internet	O
latencies	O
,	O
e.g.	O
ranging	O
from	O
40	O
to	O
80	O
milliseconds	O
round	O
trip	O
time	O
to	O
the	O
storage	O
server	O
,	O
peers	O
using	O
BuddyCache	B-KEY
can	O
reduce	O
by	O
up	O
to	O
50	O
%	O
the	O
latency	O
of	O
access	O
to	O
shared	O
objects	O
compared	O
to	O
accessing	O
the	O
remote	O
servers	O
directly	O
.	O

Addressing	O
Strategic	B-KEY
Behavior	I-KEY
in	O
a	O
Deployed	O
Microeconomic	O
Resource	B-KEY
Allocator	I-KEY
ABSTRACT	O
While	O
market-based	O
systems	O
have	O
long	O
been	O
proposed	O
as	O
solutions	O
for	O
distributed	O
resource	B-KEY
allocation	I-KEY
,	O
few	O
have	O
been	O
deployed	O
for	O
production	O
use	O
in	O
real	O
computer	O
systems	O
.	O
Towards	O
this	O
end	O
,	O
we	O
present	O
our	O
initial	O
experience	O
using	O
Mirage	O
,	O
a	O
microeconomic	O
resource	B-KEY
allocation	I-KEY
system	O
based	O
on	O
a	O
repeated	O
combinatorial	O
auction	O
.	O
Mirage	O
allocates	O
time	O
on	O
a	O
heavily-used	O
148-node	O
wireless	O
sensor	O
network	O
testbed	O
.	O
In	O
particular	O
,	O
we	O
focus	O
on	O
observed	O
strategic	O
user	O
behavior	O
over	O
a	O
four-month	O
period	O
in	O
which	O
312,148	O
node	O
hours	O
were	O
allocated	O
across	O
11	O
research	O
projects	O
.	O
Based	O
on	O
these	O
results	O
,	O
we	O
present	O
a	O
set	O
of	O
key	O
challenges	O
for	O
market-based	O
resource	B-KEY
allocation	I-KEY
systems	O
based	O
on	O
repeated	O
combinatorial	O
auctions	O
.	O
Finally	O
,	O
we	O
propose	O
refinements	O
to	O
the	O
system	O
's	O
current	O
auction	O
scheme	O
to	O
mitigate	O
the	O
strategies	O
observed	O
to	O
date	O
and	O
also	O
comment	O
on	O
some	O
initial	O
steps	O
toward	O
building	O
an	O
approximately	O
strategyproof	O
repeated	O
combinatorial	B-KEY
auction	I-KEY
.	O

Betting	B-KEY
Boolean-Style	O
:	O
A	O
Framework	O
for	O
Trading	O
in	O
Securities	O
Based	O
on	O
Logical	O
Formulas	O
ABSTRACT	O
We	O
develop	O
a	O
framework	O
for	O
trading	O
in	O
compound	B-KEY
securities	I-KEY
:	O
financial	O
instruments	O
that	O
pay	O
off	O
contingent	O
on	O
the	O
outcomes	O
of	O
arbitrary	O
statements	O
in	O
propositional	O
logic	O
.	O
Buying	O
or	O
selling	O
securities	O
--	O
which	O
can	O
be	O
thought	O
of	O
as	O
betting	B-KEY
on	O
or	O
against	O
a	O
particular	O
future	O
outcome	O
--	O
allows	O
agents	O
both	O
to	O
hedge	B-KEY
risk	O
and	O
to	O
profit	O
-LRB-	O
in	O
expectation	O
-RRB-	O
on	O
subjective	O
predictions	O
.	O
A	O
compound	B-KEY
securities	I-KEY
market	O
allows	O
agents	O
to	O
place	O
bets	O
on	O
arbitrary	O
boolean	O
combinations	O
of	O
events	O
,	O
enabling	O
them	O
to	O
more	O
closely	O
achieve	O
their	O
optimal	O
risk	O
exposure	O
,	O
and	O
enabling	O
the	O
market	O
as	O
a	O
whole	O
to	O
more	O
closely	O
achieve	O
the	O
social	O
optimum	O
.	O
The	O
tradeoff	O
for	O
allowing	O
such	O
expressivity	O
is	O
in	O
the	O
complexity	O
of	O
the	O
agents	O
'	O
and	O
auctioneer	O
's	O
optimization	O
problems	O
.	O
We	O
develop	O
and	O
motivate	O
the	O
concept	O
of	O
a	O
compound	B-KEY
securities	I-KEY
market	O
,	O
presenting	O
the	O
framework	O
through	O
a	O
series	O
of	O
formal	O
definitions	O
and	O
examples	O
.	O
We	O
then	O
analyze	O
in	O
detail	O
the	O
auctioneer	O
's	O
matching	O
problem	O
.	O
We	O
show	O
that	O
,	O
with	O
n	O
events	O
,	O
the	O
matching	O
problem	O
is	O
co-NP-complete	O
in	O
the	O
divisible	O
case	O
and	O
Σp2-complete	O
in	O
the	O
indivisible	O
case	O
.	O
We	O
show	O
that	O
the	O
latter	O
hardness	O
result	O
holds	O
even	O
under	O
severe	O
language	O
restrictions	O
on	O
bids	O
.	O
With	O
log	O
n	O
events	O
,	O
the	O
problem	O
is	O
polynomial	O
in	O
the	O
divisible	O
case	O
and	O
NP-complete	O
in	O
the	O
indivisible	O
case	O
.	O
We	O
briefly	O
discuss	O
matching	O
algorithms	O
and	O
tractable	O
special	O
cases	O
.	O

Robust	O
Test	B-KEY
Collections	I-KEY
for	O
Retrieval	O
Evaluation	B-KEY
ABSTRACT	O
Low-cost	O
methods	O
for	O
acquiring	O
relevance	O
judgments	O
can	O
be	O
a	O
boon	O
to	O
researchers	O
who	O
need	O
to	O
evaluate	B-KEY
new	O
retrieval	O
tasks	O
or	O
topics	O
but	O
do	O
not	O
have	O
the	O
resources	O
to	O
make	O
thousands	O
of	O
judgments	O
.	O
While	O
these	O
judgments	O
are	O
very	O
useful	O
for	O
a	O
one-time	O
evaluation	B-KEY
,	O
it	O
is	O
not	O
clear	O
that	O
they	O
can	O
be	O
trusted	O
when	O
re-used	O
to	O
evaluate	B-KEY
new	O
systems	O
.	O
In	O
this	O
work	O
,	O
we	O
formally	O
define	O
what	O
it	O
means	O
for	O
judgments	O
to	O
be	O
reusable	B-KEY
:	O
the	O
confidence	O
in	O
an	O
evaluation	B-KEY
of	O
new	O
systems	O
can	O
be	O
accurately	O
assessed	O
from	O
an	O
existing	O
set	O
of	O
relevance	O
judgments	O
.	O
We	O
then	O
present	O
a	O
method	O
for	O
augmenting	O
a	O
set	O
of	O
relevance	O
judgments	O
with	O
relevance	O
estimates	O
that	O
require	O
no	O
additional	O
assessor	O
effort	O
.	O
Using	O
this	O
method	O
practically	O
guarantees	O
reusability	B-KEY
:	O
with	O
as	O
few	O
as	O
five	O
judgments	O
per	O
topic	O
taken	O
from	O
only	O
two	O
systems	O
,	O
we	O
can	O
reliably	O
evaluate	B-KEY
a	O
larger	O
set	O
of	O
ten	O
systems	O
.	O
Even	O
the	O
smallest	O
sets	O
of	O
judgments	O
can	O
be	O
useful	O
for	O
evaluation	B-KEY
of	O
new	O
systems	O
.	O

Investigating	O
the	O
Querying	B-KEY
and	O
Browsing	O
Behavior	O
of	O
Advanced	O
Search	B-KEY
Engine	I-KEY
Users	O
ABSTRACT	O
One	O
way	O
to	O
help	O
all	O
users	O
of	O
commercial	O
Web	O
search	B-KEY
engines	I-KEY
be	O
more	O
successful	O
in	O
their	O
searches	O
is	O
to	O
better	O
understand	O
what	O
those	O
users	O
with	O
greater	O
search	O
expertise	O
are	O
doing	O
,	O
and	O
use	O
this	O
knowledge	O
to	O
benefit	O
everyone	O
.	O
In	O
this	O
paper	O
we	O
study	O
the	O
interaction	O
logs	O
of	O
advanced	O
search	B-KEY
engine	I-KEY
users	O
-LRB-	O
and	O
those	O
not	O
so	O
advanced	O
-RRB-	O
to	O
better	O
understand	O
how	O
these	O
user	O
groups	O
search	O
.	O
The	O
results	O
show	O
that	O
there	O
are	O
marked	O
differences	O
in	O
the	O
queries	B-KEY
,	O
result	O
clicks	O
,	O
post-query	O
browsing	O
,	O
and	O
search	O
success	O
of	O
users	O
we	O
classify	O
as	O
advanced	O
-LRB-	O
based	O
on	O
their	O
use	O
of	O
query	O
operators	O
-RRB-	O
,	O
relative	O
to	O
those	O
classified	O
as	O
non-advanced	O
.	O
Our	O
findings	O
have	O
implications	O
for	O
how	O
advanced	O
users	O
should	O
be	O
supported	O
during	O
their	O
searches	O
,	O
and	O
how	O
their	O
interactions	O
could	O
be	O
used	O
to	O
help	O
searchers	O
of	O
all	O
experience	O
levels	O
find	O
more	O
relevant	B-KEY
information	O
and	O
learn	O
improved	O
searching	O
strategies	O
.	O

Dynamic	B-KEY
Semantics	I-KEY
for	O
Agent	B-KEY
Communication	I-KEY
Languages	I-KEY
ABSTRACT	O
This	O
paper	O
proposes	O
dynamic	B-KEY
semantics	I-KEY
for	O
agent	B-KEY
communication	I-KEY
languages	I-KEY
-LRB-	O
ACLs	O
-RRB-	O
as	O
a	O
method	O
for	O
tackling	O
some	O
of	O
the	O
fundamental	O
problems	O
associated	O
with	O
agent	O
communication	O
in	O
open	O
multiagent	O
systems	O
.	O
Based	O
on	O
the	O
idea	O
of	O
providing	O
alternative	O
semantic	O
``	O
variants	O
''	O
for	O
speech	O
acts	O
and	O
transition	O
rules	O
between	O
them	O
that	O
are	O
contingent	O
on	O
previous	O
agent	O
behaviour	O
,	O
our	O
framework	O
provides	O
an	O
improved	O
notion	O
of	O
grounding	O
semantics	O
in	O
ongoing	O
interaction	O
,	O
a	O
simple	O
mechanism	O
for	O
distinguishing	O
between	O
compliant	O
and	O
expected	O
behaviour	O
,	O
and	O
a	O
way	O
to	O
specify	O
sanction	O
and	O
reward	O
mechanisms	O
as	O
part	O
of	O
the	O
ACL	O
itself	O
.	O
We	O
extend	O
a	O
common	O
framework	O
for	O
commitment-based	O
ACL	O
semantics	O
to	O
obtain	O
these	O
properties	O
,	O
discuss	O
desiderata	O
for	O
the	O
design	O
of	O
concrete	O
dynamic	B-KEY
semantics	I-KEY
together	O
with	O
examples	O
,	O
and	O
analyse	O
their	O
properties	O
.	O

Runtime	O
Metrics	B-KEY
Collection	I-KEY
for	O
Middleware	O
Supported	O
Adaptation	B-KEY
of	O
Mobile	O
Applications	O
ABSTRACT	O
This	O
paper	O
proposes	O
,	O
implements	O
,	O
and	O
evaluates	O
in	O
terms	O
of	O
worst	O
case	O
performance	O
,	O
an	O
online	O
metrics	B-KEY
collection	I-KEY
strategy	O
to	O
facilitate	O
application	O
adaptation	B-KEY
via	O
object	O
mobility	O
using	O
a	O
mobile	B-KEY
object	I-KEY
framework	O
and	O
supporting	O
middleware	O
.	O
The	O
solution	O
is	O
based	O
upon	O
an	O
abstract	O
representation	O
of	O
the	O
mobile	B-KEY
object	I-KEY
system	O
,	O
which	O
holds	O
containers	O
aggregating	O
metrics	O
for	O
each	O
specific	O
component	O
including	O
host	O
managers	O
,	O
runtimes	O
and	O
mobile	B-KEY
objects	I-KEY
.	O
A	O
key	O
feature	O
of	O
the	O
solution	O
is	O
the	O
specification	O
of	O
multiple	O
configurable	O
criteria	O
to	O
control	O
the	O
measurement	B-KEY
and	O
propagation	O
of	O
metrics	O
through	O
the	O
system	O
.	O
The	O
MobJeX	B-KEY
platform	O
was	O
used	O
as	O
the	O
basis	O
for	O
implementation	O
and	O
testing	O
with	O
a	O
number	O
of	O
laboratory	O
tests	O
conducted	O
to	O
measure	B-KEY
scalability	O
,	O
efficiency	O
and	O
the	O
application	O
of	O
simple	O
measurement	B-KEY
and	O
propagation	O
criteria	O
to	O
reduce	O
collection	O
overhead	O
.	O

Understanding	O
User	O
Behavior	O
in	O
Online	O
Feedback	O
Reporting	O
ABSTRACT	O
Online	B-KEY
reviews	I-KEY
have	O
become	O
increasingly	O
popular	O
as	O
a	O
way	O
to	O
judge	O
the	O
quality	O
of	O
various	O
products	O
and	O
services	O
.	O
Previous	O
work	O
has	O
demonstrated	O
that	O
contradictory	O
reporting	O
and	O
underlying	O
user	O
biases	O
make	O
judging	O
the	O
true	O
worth	O
of	O
a	O
service	O
difficult	O
.	O
In	O
this	O
paper	O
,	O
we	O
investigate	O
underlying	O
factors	O
that	O
influence	O
user	O
behavior	O
when	O
reporting	O
feedback	O
.	O
We	O
look	O
at	O
two	O
sources	O
of	O
information	O
besides	O
numerical	O
ratings	B-KEY
:	O
linguistic	O
evidence	O
from	O
the	O
textual	O
comment	O
accompanying	O
a	O
review	O
,	O
and	O
patterns	O
in	O
the	O
time	O
sequence	O
of	O
reports	O
.	O
We	O
first	O
show	O
that	O
groups	O
of	O
users	O
who	O
amply	O
discuss	O
a	O
certain	O
feature	O
are	O
more	O
likely	O
to	O
agree	O
on	O
a	O
common	O
rating	B-KEY
for	O
that	O
feature	O
.	O
Second	O
,	O
we	O
show	O
that	O
a	O
user	O
's	O
rating	B-KEY
partly	O
reflects	O
the	O
difference	O
between	O
true	O
quality	O
and	O
prior	O
expectation	O
of	O
quality	O
as	O
inferred	O
from	O
previous	O
reviews	O
.	O
Both	O
give	O
us	O
a	O
less	O
noisy	O
way	O
to	O
produce	O
rating	B-KEY
estimates	O
and	O
reveal	O
the	O
reasons	O
behind	O
user	O
bias	O
.	O
Our	O
hypotheses	O
were	O
validated	O
by	O
statistical	O
evidence	O
from	O
hotel	O
reviews	O
on	O
the	O
TripAdvisor	O
website	O
.	O

Temporal	O
Linear	B-KEY
Logic	I-KEY
as	O
a	O
Basis	O
for	O
Flexible	O
Agent	O
Interactions	O
ABSTRACT	O
Interactions	O
between	O
agents	O
in	O
an	O
open	O
system	O
such	O
as	O
the	O
Internet	O
require	O
a	O
significant	O
degree	O
of	O
flexibility	O
.	O
A	O
crucial	O
aspect	O
of	O
the	O
development	O
of	O
such	O
methods	O
is	O
the	O
notion	O
of	O
commitments	O
,	O
which	O
provides	O
a	O
mechanism	O
for	O
coordinating	O
interactive	B-KEY
behaviors	I-KEY
among	O
agents	O
.	O
In	O
this	O
paper	O
,	O
we	O
investigate	O
an	O
approach	O
to	O
model	O
commitments	O
with	O
tight	O
integration	O
with	O
protocol	O
actions	O
.	O
This	O
means	O
that	O
there	O
is	O
no	O
need	O
to	O
have	O
an	O
explicit	O
mapping	O
from	O
protocols	O
actions	O
to	O
operations	O
on	O
commitments	O
and	O
an	O
external	O
mechanism	O
to	O
process	O
and	O
enforce	O
commitments	O
.	O
We	O
show	O
how	O
agents	O
can	O
reason	O
about	O
commitments	O
and	O
protocol	O
actions	O
to	O
achieve	O
the	O
end	O
results	O
of	O
protocols	O
using	O
a	O
reasoning	O
system	O
based	O
on	O
temporal	O
linear	B-KEY
logic	I-KEY
,	O
which	O
incorporates	O
both	O
temporal	O
and	O
resource-sensitive	O
reasoning	O
.	O
We	O
also	O
discuss	O
the	O
application	O
of	O
this	O
framework	O
to	O
scenarios	O
such	O
as	O
online	O
commerce	O
.	O

The	O
Role	O
of	O
Compatibility	O
in	O
the	O
Diffusion	O
of	O
Technologies	O
Through	O
Social	O
Networks	O
ABSTRACT	O
In	O
many	O
settings	O
,	O
competing	O
technologies	O
--	O
for	O
example	O
,	O
operating	O
systems	O
,	O
instant	O
messenger	O
systems	O
,	O
or	O
document	O
formats	O
--	O
can	O
be	O
seen	O
adopting	O
a	O
limited	O
amount	O
of	O
compatibility	O
with	O
one	O
another	O
;	O
in	O
other	O
words	O
,	O
the	O
difficulty	O
in	O
using	O
multiple	O
technologies	O
is	O
balanced	O
somewhere	O
between	O
the	O
two	O
extremes	O
of	O
impossibility	O
and	O
effortless	O
interoperability	B-KEY
.	O
There	O
are	O
a	O
range	O
of	O
reasons	O
why	O
this	O
phenomenon	O
occurs	O
,	O
many	O
of	O
which	O
--	O
based	O
on	O
legal	O
,	O
social	O
,	O
or	O
business	O
considerations	O
--	O
seem	O
to	O
defy	O
concise	O
mathematical	O
models	O
.	O
Despite	O
this	O
,	O
we	O
show	O
that	O
the	O
advantages	O
of	O
limited	B-KEY
compatibility	I-KEY
can	O
arise	O
in	O
a	O
very	O
simple	O
model	O
of	O
diffusion	O
in	O
social	O
networks	O
,	O
thus	O
offering	O
a	O
basic	O
explanation	O
for	O
this	O
phenomenon	O
in	O
purely	O
strategic	O
terms	O
.	O
Our	O
approach	O
builds	O
on	O
work	O
on	O
the	O
diffusion	B-KEY
of	I-KEY
innovations	I-KEY
in	O
the	O
economics	O
literature	O
,	O
which	O
seeks	O
to	O
model	O
how	O
a	O
new	O
technology	O
A	O
might	O
spread	O
through	O
a	O
social	O
network	O
of	O
individuals	O
who	O
are	O
currently	O
users	O
of	O
technology	O
B	O
.	O
We	O
consider	O
several	O
ways	O
of	O
capturing	O
the	O
compatibility	O
of	O
A	O
and	O
B	O
,	O
focusing	O
primarily	O
on	O
a	O
model	O
in	O
which	O
users	O
can	O
choose	O
to	O
adopt	O
A	O
,	O
adopt	O
B	O
,	O
or	O
--	O
at	O
an	O
extra	O
cost	O
--	O
adopt	O
both	O
A	O
and	O
B	O
.	O
We	O
characterize	B-KEY
how	O
the	O
ability	O
of	O
A	O
to	O
spread	O
depends	O
on	O
both	O
its	O
quality	O
relative	O
to	O
B	O
,	O
and	O
also	O
this	O
additional	O
cost	O
of	O
adopting	O
both	O
,	O
and	O
find	O
some	O
surprising	O
non-monotonicity	O
properties	O
in	O
the	O
dependence	O
on	O
these	O
parameters	O
:	O
in	O
some	O
cases	O
,	O
for	O
one	O
technology	O
to	O
survive	O
the	O
introduction	O
of	O
another	O
,	O
the	O
cost	O
of	O
adopting	O
both	O
technologies	O
must	O
be	O
balanced	O
within	O
a	O
narrow	O
,	O
intermediate	O
range	O
.	O
We	O
also	O
extend	O
the	O
framework	O
to	O
the	O
case	O
of	O
multiple	O
technologies	O
,	O
where	O
we	O
find	O
that	O
a	O
simple	O
This	O
work	O
has	O
been	O
supported	O
in	O
part	O
by	O
NSF	O
grants	O
CCF0325453	O
,	O
IIS-0329064	O
,	O
CNS-0403340	O
,	O
and	O
BCS-0537606	O
,	O
a	O
Google	O
Research	O
Grant	O
,	O
a	O
Yahoo!	O
Research	O
Alliance	O
Grant	O
,	O
the	O
Institute	O
for	O
the	O
Social	O
Sciences	O
at	O
Cornell	O
,	O
and	O
the	O
John	O
D.	O
and	O
Catherine	O
T.	O
MacArthur	O
Foundation	O
.	O
model	O
captures	O
the	O
phenomenon	O
of	O
two	O
firms	O
adopting	O
a	O
limited	O
``	O
strategic	O
alliance	O
''	O
to	O
defend	O
against	O
a	O
new	O
,	O
third	O
technology	O
.	O

Sensor	O
Deployment	B-KEY
Strategy	O
for	O
Target	B-KEY
Detection	I-KEY
ABSTRACT	O
In	O
order	O
to	O
monitor	O
a	O
region	O
for	O
traffic	O
traversal	O
,	O
sensors	O
can	O
be	O
deployed	B-KEY
to	O
perform	O
collaborative	B-KEY
target	I-KEY
detection	I-KEY
.	O
Such	O
a	O
sensor	B-KEY
network	I-KEY
achieves	O
a	O
certain	O
level	O
of	O
detection	O
performance	O
with	O
an	O
associated	O
cost	O
of	O
deployment	B-KEY
.	O
This	O
paper	O
addresses	O
this	O
problem	O
by	O
proposing	O
path	B-KEY
exposure	I-KEY
as	O
a	O
measure	O
of	O
the	O
goodness	O
of	O
a	O
deployment	O
and	O
presents	O
an	O
approach	O
for	O
sequential	O
deployment	O
in	O
steps	O
.	O
It	O
illustrates	O
that	O
the	O
cost	O
of	O
deployment	B-KEY
can	O
be	O
minimized	O
to	O
achieve	O
the	O
desired	O
detection	O
performance	O
by	O
appropriately	O
choosing	O
the	O
number	B-KEY
of	I-KEY
sensors	I-KEY
deployed	B-KEY
in	O
each	O
step	O
.	O

Live	O
Data	B-KEY
Center	I-KEY
Migration	I-KEY
across	O
WANs	B-KEY
:	O
A	O
Robust	O
Cooperative	O
Context	O
Aware	O
Approach	O
ABSTRACT	O
A	O
significant	O
concern	O
for	O
Internet-based	O
service	O
providers	O
is	O
the	O
continued	O
operation	O
and	O
availability	O
of	O
services	O
in	O
the	O
face	O
of	O
outages	O
,	O
whether	O
planned	O
or	O
unplanned	O
.	O
In	O
this	O
paper	O
we	O
advocate	O
a	O
cooperative	O
,	O
context-aware	O
approach	O
to	O
data	B-KEY
center	I-KEY
migration	I-KEY
across	O
WANs	B-KEY
to	O
deal	O
with	O
outages	O
in	O
a	O
non-disruptive	O
manner	O
.	O
We	O
specifically	O
seek	O
to	O
achieve	O
high	O
availability	O
of	O
data	O
center	O
services	O
in	O
the	O
face	O
of	O
both	O
planned	O
and	O
unanticipated	O
outages	O
of	O
data	O
center	O
facilities	O
.	O
We	O
make	O
use	O
of	O
server	O
virtualization	O
technologies	O
to	O
enable	O
the	O
replication	O
and	O
migration	O
of	O
server	O
functions	O
.	O
We	O
propose	O
new	O
network	O
functions	O
to	O
enable	O
server	O
migration	O
and	O
replication	O
across	O
wide	O
area	O
networks	O
-LRB-	O
e.g.	O
,	O
the	O
Internet	O
-RRB-	O
,	O
and	O
finally	O
show	O
the	O
utility	O
of	O
intelligent	O
and	O
dynamic	O
storage	B-KEY
replication	O
technology	O
to	O
ensure	O
applications	O
have	O
access	O
to	O
data	O
in	O
the	O
face	O
of	O
outages	O
with	O
very	O
tight	O
recovery	O
point	O
objectives	O
.	O

Real-Time	O
Agent	O
Characterization	O
and	O
Prediction	B-KEY
ABSTRACT	O
Reasoning	O
about	O
agents	O
that	O
we	O
observe	O
in	O
the	O
world	O
is	O
challenging	O
.	O
Our	O
available	O
information	O
is	O
often	O
limited	O
to	O
observations	O
of	O
the	O
agent	O
's	O
external	B-KEY
behavior	I-KEY
in	O
the	O
past	O
and	O
present	O
.	O
To	O
understand	O
these	O
actions	O
,	O
we	O
need	O
to	O
deduce	O
the	O
agent	O
's	O
internal	B-KEY
state	I-KEY
,	O
which	O
includes	O
not	O
only	O
rational	O
elements	O
-LRB-	O
such	O
as	O
intentions	O
and	O
plans	O
-RRB-	O
,	O
but	O
also	O
emotive	B-KEY
ones	O
-LRB-	O
such	O
as	O
fear	O
-RRB-	O
.	O
In	O
addition	O
,	O
we	O
often	O
want	O
to	O
predict	B-KEY
the	O
agent	O
's	O
future	O
actions	O
,	O
which	O
are	O
constrained	O
not	O
only	O
by	O
these	O
inward	O
characteristics	O
,	O
but	O
also	O
by	O
the	O
dynamics	B-KEY
of	O
the	O
agent	O
's	O
interaction	O
with	O
its	O
environment	O
.	O
BEE	O
-LRB-	O
Behavior	O
Evolution	B-KEY
and	O
Extrapolation	O
-RRB-	O
uses	O
a	O
faster-than-real-time	O
agentbased	O
model	O
of	O
the	O
environment	O
to	O
characterize	O
agents	O
'	O
internal	O
state	O
by	O
evolution	O
against	O
observed	O
behavior	O
,	O
and	O
then	O
predict	O
their	O
future	O
behavior	O
,	O
taking	O
into	O
account	O
the	O
dynamics	O
of	O
their	O
interaction	O
with	O
the	O
environment	O
.	O

A	O
Support	O
Vector	O
Method	O
for	O
Optimizing	O
Average	O
Precision	O
ABSTRACT	O
Machine	B-KEY
learning	I-KEY
is	O
commonly	O
used	O
to	O
improve	O
ranked	B-KEY
retrieval	O
systems	O
.	O
Due	O
to	O
computational	O
difficulties	O
,	O
few	O
learning	B-KEY
techniques	I-KEY
have	O
been	O
developed	O
to	O
directly	O
optimize	O
for	O
mean	B-KEY
average	I-KEY
precision	I-KEY
-LRB-	O
MAP	O
-RRB-	O
,	O
despite	O
its	O
widespread	O
use	O
in	O
evaluating	O
such	O
systems	O
.	O
Existing	O
approaches	O
optimizing	O
MAP	O
either	O
do	O
not	O
find	O
a	O
globally	O
optimal	B-KEY
solution	I-KEY
,	O
or	O
are	O
computationally	O
expensive	O
.	O
In	O
contrast	O
,	O
we	O
present	O
a	O
general	O
SVM	O
learning	O
algorithm	O
that	O
efficiently	O
finds	O
a	O
globally	O
optimal	B-KEY
solution	I-KEY
to	O
a	O
straightforward	O
relaxation	B-KEY
of	I-KEY
MAP	I-KEY
.	O
We	O
evaluate	O
our	O
approach	O
using	O
the	O
TREC	O
9	O
and	O
TREC	O
10	O
Web	O
Track	O
corpora	O
-LRB-	O
WT10g	O
-RRB-	O
,	O
comparing	O
against	O
SVMs	O
optimized	O
for	O
accuracy	O
and	O
ROCArea	O
.	O
In	O
most	O
cases	O
we	O
show	O
our	O
method	O
to	O
produce	O
statistically	O
significant	O
improvements	O
in	O
MAP	O
scores	O
.	O

Encryption-Enforced	O
Access	O
Control	O
in	O
Dynamic	O
Multi-Domain	B-KEY
Publish/Subscribe	O
Networks	O
ABSTRACT	O
Publish/subscribe	O
systems	O
provide	O
an	O
efficient	O
,	O
event-based	O
,	O
wide-area	O
distributed	O
communications	O
infrastructure	O
.	O
Large	O
scale	O
publish/subscribe	O
systems	O
are	O
likely	O
to	O
employ	O
components	O
of	O
the	O
event	O
transport	O
network	O
owned	O
by	O
cooperating	O
,	O
but	O
independent	O
organisations	O
.	O
As	O
the	O
number	O
of	O
participants	O
in	O
the	O
network	O
increases	O
,	O
security	O
becomes	O
an	O
increasing	O
concern	O
.	O
This	O
paper	O
extends	O
previous	O
work	O
to	O
present	O
and	O
evaluate	O
a	O
secure	O
multi-domain	B-KEY
publish/subscribe	O
infrastructure	O
that	O
supports	O
and	O
enforces	O
fine-grained	O
access	O
control	O
over	O
the	O
individual	O
attributes	O
of	O
event	O
types	O
.	O
Key	O
refresh	O
allows	O
us	O
to	O
ensure	O
forward	O
and	O
backward	O
security	O
when	O
event	O
brokers	O
join	O
and	O
leave	O
the	O
network	O
.	O
We	O
demonstrate	O
that	O
the	O
time	O
and	O
space	O
overheads	O
can	O
be	O
minimised	O
by	O
careful	O
consideration	O
of	O
encryption	B-KEY
techniques	O
,	O
and	O
by	O
the	O
use	O
of	O
caching	O
to	O
decrease	O
unnecessary	O
decryptions	O
.	O
We	O
show	O
that	O
our	O
approach	O
has	O
a	O
smaller	O
overall	B-KEY
communication	I-KEY
overhead	I-KEY
than	O
existing	O
approaches	O
for	O
achieving	O
the	O
same	O
degree	O
of	O
control	O
over	O
security	O
in	O
publish/subscribe	O
networks	O
.	O

A	O
Framework	O
for	O
Architecting	O
Peer-to-Peer	O
Receiver-driven	O
Overlays	O
ABSTRACT	O
This	O
paper	O
presents	O
a	O
simple	O
and	O
scalable	O
framework	O
for	O
architecting	O
peer-to-peer	O
overlays	O
called	O
Peer-to-peer	O
Receiverdriven	O
Overlay	O
-LRB-	O
or	O
PRO	B-KEY
-RRB-	O
.	O
PRO	B-KEY
is	O
designed	B-KEY
for	O
non-interactive	O
streaming	O
applications	O
and	O
its	O
primary	O
design	B-KEY
goal	O
is	O
to	O
maximize	O
delivered	O
bandwidth	O
-LRB-	O
and	O
thus	O
delivered	O
quality	O
-RRB-	O
to	O
peers	O
with	O
heterogeneous	O
and	O
asymmetric	O
bandwidth	O
.	O
To	O
achieve	O
this	O
goal	O
,	O
PRO	B-KEY
adopts	O
a	O
receiver-driven	O
approach	O
where	O
each	O
receiver	O
-LRB-	O
or	O
participating	O
peer	O
-RRB-	O
-LRB-	O
i	O
-RRB-	O
independently	O
discovers	O
other	O
peers	O
in	O
the	O
overlay	O
through	O
gossiping	O
,	O
and	O
-LRB-	O
ii	O
-RRB-	O
selfishly	O
determines	O
the	O
best	O
subset	O
of	O
parent	O
peers	O
through	O
which	O
to	O
connect	O
to	O
the	O
overlay	O
to	O
maximize	O
its	O
own	O
delivered	O
bandwidth	O
.	O
Participating	O
peers	O
form	O
an	O
unstructured	O
overlay	O
which	O
is	O
inherently	O
robust	O
to	O
high	O
churn	O
rate	O
.	O
Furthermore	O
,	O
each	O
receiver	O
leverages	O
congestion	B-KEY
controlled	I-KEY
bandwidth	O
from	O
its	O
parents	O
as	O
implicit	O
signal	O
to	O
detect	O
and	O
react	O
to	O
long-term	O
changes	O
in	O
network	O
or	O
overlay	O
condition	O
without	O
any	O
explicit	O
coordination	O
with	O
other	O
participating	O
peers	O
.	O
Independent	O
parent	O
selection	O
by	O
individual	O
peers	O
dynamically	O
converge	O
to	O
an	O
efficient	O
overlay	O
structure	O
.	O

A	O
Study	O
of	O
Poisson	O
Query	B-KEY
Generation	I-KEY
Model	O
for	O
Information	O
Retrieval	O
ABSTRACT	O
Many	O
variants	O
of	O
language	B-KEY
models	I-KEY
have	O
been	O
proposed	O
for	O
information	O
retrieval	O
.	O
Most	O
existing	O
models	O
are	O
based	O
on	O
multinomial	B-KEY
distribution	I-KEY
and	O
would	O
score	O
documents	O
based	O
on	O
query	O
likelihood	O
computed	O
based	O
on	O
a	O
query	B-KEY
generation	I-KEY
probabilistic	O
model	O
.	O
In	O
this	O
paper	O
,	O
we	O
propose	O
and	O
study	O
a	O
new	O
family	O
of	O
query	B-KEY
generation	I-KEY
models	O
based	O
on	O
Poisson	B-KEY
distribution	I-KEY
.	O
We	O
show	O
that	O
while	O
in	O
their	O
simplest	O
forms	O
,	O
the	O
new	O
family	O
of	O
models	O
and	O
the	O
existing	O
multinomial	O
models	O
are	O
equivalent	O
,	O
they	O
behave	O
differently	O
for	O
many	O
smoothing	O
methods	O
.	O
We	O
show	O
that	O
the	O
Poisson	O
model	O
has	O
several	O
advantages	O
over	O
the	O
multinomial	O
model	O
,	O
including	O
naturally	O
accommodating	O
per-term	O
smoothing	O
and	O
allowing	O
for	O
more	O
accurate	O
background	O
modeling	O
.	O
We	O
present	O
several	O
variants	O
of	O
the	O
new	O
model	O
corresponding	O
to	O
different	O
smoothing	O
methods	O
,	O
and	O
evaluate	O
them	O
on	O
four	O
representative	O
TREC	O
test	O
collections	O
.	O
The	O
results	O
show	O
that	O
while	O
their	O
basic	O
models	O
perform	O
comparably	O
,	O
the	O
Poisson	O
model	O
can	O
outperform	O
multinomial	O
model	O
with	O
per-term	O
smoothing	O
.	O
The	O
performance	O
can	O
be	O
further	O
improved	O
with	O
two-stage	O
smoothing	O
.	O

Personalized	O
Query	B-KEY
Expansion	I-KEY
for	O
the	O
Web	O
ABSTRACT	O
The	O
inherent	O
ambiguity	O
of	O
short	B-KEY
keyword	I-KEY
queries	I-KEY
demands	O
for	O
enhanced	O
methods	O
for	O
Web	B-KEY
retrieval	I-KEY
.	O
In	O
this	O
paper	O
we	O
propose	O
to	O
improve	O
such	O
Web	B-KEY
queries	I-KEY
by	O
expanding	O
them	O
with	O
terms	O
collected	O
from	O
each	O
user	O
's	O
Personal	B-KEY
Information	I-KEY
Repository	I-KEY
,	O
thus	O
implicitly	O
personalizing	O
the	O
search	B-KEY
output	I-KEY
.	O
We	O
introduce	O
five	O
broad	O
techniques	O
for	O
generating	O
the	O
additional	B-KEY
query	I-KEY
keywords	I-KEY
by	O
analyzing	O
user	O
data	O
at	O
increasing	O
granularity	B-KEY
levels	I-KEY
,	O
ranging	O
from	O
term	B-KEY
and	I-KEY
compound	I-KEY
level	I-KEY
analysis	I-KEY
up	O
to	O
global	B-KEY
co-occurrence	I-KEY
statistics	I-KEY
,	O
as	O
well	O
as	O
to	O
using	O
external	O
thesauri	O
.	O
Our	O
extensive	B-KEY
empirical	I-KEY
analysis	I-KEY
under	O
four	O
different	O
scenarios	O
shows	O
some	O
of	O
these	O
approaches	O
to	O
perform	O
very	O
well	O
,	O
especially	O
on	O
ambiguous	B-KEY
queries	I-KEY
,	O
producing	O
a	O
very	O
strong	O
increase	O
in	O
the	O
quality	B-KEY
of	O
the	O
output	B-KEY
rankings	I-KEY
.	O
Subsequently	O
,	O
we	O
move	O
this	O
personalized	B-KEY
search	I-KEY
framework	I-KEY
one	O
step	O
further	O
and	O
propose	O
to	O
make	O
the	O
expansion	B-KEY
process	I-KEY
adaptive	O
to	O
various	B-KEY
features	I-KEY
of	I-KEY
each	I-KEY
query	I-KEY
.	O
A	O
separate	O
set	O
of	O
experiments	O
indicates	O
the	O
adaptive	B-KEY
algorithms	I-KEY
to	O
bring	O
an	O
additional	O
statistically	O
significant	B-KEY
improvement	I-KEY
over	O
the	O
best	O
static	B-KEY
expansion	I-KEY
approach	I-KEY
.	O

Efficient	O
Bayesian	O
Hierarchical	O
User	O
Modeling	B-KEY
for	O
Recommendation	B-KEY
Systems	I-KEY
ABSTRACT	O
A	O
content-based	O
personalized	B-KEY
recommendation	B-KEY
system	I-KEY
learns	O
user	O
specific	O
profiles	O
from	O
user	O
feedback	O
so	O
that	O
it	O
can	O
deliver	O
information	O
tailored	O
to	O
each	O
individual	O
user	O
's	O
interest	O
.	O
A	O
system	O
serving	O
millions	O
of	O
users	O
can	O
learn	O
a	O
better	O
user	O
profile	O
for	O
a	O
new	O
user	O
,	O
or	O
a	O
user	O
with	O
little	O
feedback	O
,	O
by	O
borrowing	O
information	O
from	O
other	O
users	O
through	O
the	O
use	O
of	O
a	O
Bayesian	B-KEY
hierarchical	I-KEY
model	I-KEY
.	O
Learning	O
the	O
model	B-KEY
parameters	B-KEY
to	O
optimize	O
the	O
joint	O
data	O
likelihood	O
from	O
millions	O
of	O
users	O
is	O
very	O
computationally	O
expensive	O
.	O
The	O
commonly	O
used	O
EM	B-KEY
algorithm	I-KEY
converges	O
very	O
slowly	O
due	O
to	O
the	O
sparseness	O
of	O
the	O
data	O
in	O
IR	B-KEY
applications	O
.	O
This	O
paper	O
proposes	O
a	O
new	O
fast	O
learning	B-KEY
technique	I-KEY
to	O
learn	O
a	O
large	O
number	O
of	O
individual	O
user	O
profiles	O
.	O
The	O
efficacy	O
and	O
efficiency	O
of	O
the	O
proposed	O
algorithm	O
are	O
justified	O
by	O
theory	O
and	O
demonstrated	O
on	O
actual	O
user	O
data	O
from	O
Netflix	O
and	O
MovieLens	O
.	O

Sharing	O
Experiences	O
to	O
Learn	O
User	O
Characteristics	O
in	O
Dynamic	O
Environments	O
with	O
Sparse	O
Data	O
ABSTRACT	O
This	O
paper	O
investigates	O
the	O
problem	O
of	O
estimating	O
the	O
value	O
of	O
probabilistic	B-KEY
parameters	I-KEY
needed	O
for	O
decision	B-KEY
making	I-KEY
in	O
environments	O
in	O
which	O
an	O
agent	B-KEY
,	O
operating	O
within	O
a	O
multi-agent	O
system	O
,	O
has	O
no	O
a	O
priori	O
information	O
about	O
the	O
structure	O
of	O
the	O
distribution	O
of	O
parameter	O
values	O
.	O
The	O
agent	B-KEY
must	O
be	O
able	O
to	O
produce	O
estimations	O
even	O
when	O
it	O
may	O
have	O
made	O
only	O
a	O
small	O
number	O
of	O
direct	O
observations	O
,	O
and	O
thus	O
it	O
must	O
be	O
able	O
to	O
operate	O
with	O
sparse	O
data	O
.	O
The	O
paper	O
describes	O
a	O
mechanism	O
that	O
enables	O
the	O
agent	B-KEY
to	O
significantly	O
improve	O
its	O
estimation	O
by	O
augmenting	O
its	O
direct	O
observations	O
with	O
those	O
obtained	O
by	O
other	O
agents	B-KEY
with	O
which	O
it	O
is	O
coordinating	O
.	O
To	O
avoid	O
undesirable	O
bias	O
in	O
relatively	O
heterogeneous	O
environments	O
while	O
effectively	O
using	O
relevant	O
data	O
to	O
improve	O
its	O
estimations	O
,	O
the	O
mechanism	O
weighs	O
the	O
contributions	O
of	O
other	O
agents	B-KEY
'	O
observations	O
based	O
on	O
a	O
real-time	O
estimation	O
of	O
the	O
level	O
of	O
similarity	O
between	O
each	O
of	O
these	O
agents	B-KEY
and	O
itself	O
.	O
The	O
``	O
coordination	O
autonomy	O
''	O
module	O
of	O
a	O
coordination-manager	O
system	O
provided	O
an	O
empirical	O
setting	O
for	O
evaluation	O
.	O
Simulation-based	O
evaluations	O
demonstrated	O
that	O
the	O
proposed	O
mechanism	O
outperforms	O
estimations	O
based	O
exclusively	O
on	O
an	O
agent	B-KEY
's	O
own	O
observations	O
as	O
well	O
as	O
estimations	O
based	O
on	O
an	O
unweighted	O
aggregate	O
of	O
all	O
other	O
agents	B-KEY
'	O
observations	O
.	O

Researches	O
on	O
Scheme	O
of	O
Pairwise	O
Key	O
Establishment	O
for	O
Distributed	O
Sensor	B-KEY
Networks	I-KEY
ABSTRACT	O
Security	B-KEY
schemes	O
of	O
pairwise	O
key	O
establishment	O
,	O
which	O
enable	O
sensors	O
to	O
communicate	O
with	O
each	O
other	O
securely	B-KEY
,	O
play	O
a	O
fundamental	O
role	O
in	O
research	O
on	O
security	B-KEY
issue	O
in	O
wireless	O
sensor	B-KEY
networks	I-KEY
.	O
A	O
new	O
kind	O
of	O
cluster	O
deployed	O
sensor	B-KEY
networks	I-KEY
distribution	O
model	O
is	O
presented	O
,	O
and	O
based	O
on	O
which	O
,	O
an	O
innovative	O
Hierarchical	B-KEY
Hypercube	I-KEY
model	I-KEY
-	O
H	O
-LRB-	O
k	O
,	O
u	O
,	O
m	O
,	O
v	O
,	O
n	O
-RRB-	O
and	O
the	O
mapping	O
relationship	O
between	O
cluster	O
deployed	O
sensor	B-KEY
networks	I-KEY
and	O
the	O
H	O
-LRB-	O
k	O
,	O
u	O
,	O
m	O
,	O
v	O
,	O
n	O
-RRB-	O
are	O
proposed	O
.	O
By	O
utilizing	O
nice	O
properties	O
of	O
H	O
-LRB-	O
k	O
,	O
u	O
,	O
m	O
,	O
v	O
,	O
n	O
-RRB-	O
model	O
,	O
a	O
new	O
general	O
framework	O
for	O
pairwise	O
key	O
predistribution	O
and	O
a	O
new	O
pairwise	O
key	O
establishment	O
algorithm	O
are	O
designed	O
,	O
which	O
combines	O
the	O
idea	O
of	O
KDC	O
-LRB-	O
Key	O
Distribution	O
Center	O
-RRB-	O
and	O
polynomial	O
pool	O
schemes	O
.	O
Furthermore	O
,	O
the	O
working	O
performance	O
of	O
the	O
newly	O
proposed	O
pairwise	O
key	O
establishment	O
algorithm	O
is	O
seriously	O
inspected	O
.	O
Theoretic	O
analysis	O
and	O
experimental	O
figures	O
show	O
that	O
the	O
new	O
algorithm	O
has	O
better	O
performance	O
and	O
provides	O
higher	O
possibilities	O
for	O
sensor	O
to	O
establish	O
pairwise	O
key	O
,	O
compared	O
with	O
previous	O
related	O
works	O
.	O

Interactions	O
between	O
Market	B-KEY
Barriers	I-KEY
and	O
Communication	O
Networks	O
in	O
Marketing	B-KEY
Systems	I-KEY
ABSTRACT	O
We	O
investigate	O
a	O
framework	O
where	O
agents	O
search	O
for	O
satisfying	O
products	O
by	O
using	O
referrals	O
from	O
other	O
agents	O
.	O
Our	O
model	O
of	O
a	O
mechanism	O
for	O
transmitting	O
word-of-mouth	O
and	O
the	O
resulting	O
behavioural	O
effects	O
is	O
based	O
on	O
integrating	O
a	O
module	O
governing	O
the	O
local	O
behaviour	O
of	O
agents	O
with	O
a	O
module	O
governing	O
the	O
structure	O
and	O
function	O
of	O
the	O
underlying	O
network	O
of	O
agents	O
.	O
Local	O
behaviour	O
incorporates	O
a	O
satisficing	O
model	O
of	O
choice	O
,	O
a	O
set	O
of	O
rules	O
governing	O
the	O
interactions	O
between	O
agents	O
,	O
including	O
learning	O
about	O
the	O
trustworthiness	O
of	O
other	O
agents	O
over	O
time	O
,	O
and	O
external	O
constraints	O
on	O
behaviour	O
that	O
may	O
be	O
imposed	O
by	O
market	B-KEY
barriers	I-KEY
or	O
switching	B-KEY
costs	I-KEY
.	O
Local	O
behaviour	O
takes	O
place	O
on	O
a	O
network	O
substrate	O
across	O
which	O
agents	O
exchange	O
positive	O
and	O
negative	O
information	O
about	O
products	O
.	O
We	O
use	O
various	O
degree	O
distributions	O
dictating	O
the	O
extent	O
of	O
connectivity	O
,	O
and	O
incorporate	O
both	O
small-world	O
effects	O
and	O
the	O
notion	O
of	O
preferential	O
attachment	O
in	O
our	O
network	O
models	O
.	O
We	O
compare	O
the	O
effectiveness	O
of	O
referral	B-KEY
systems	I-KEY
over	O
various	O
network	O
structures	O
for	O
easy	O
and	O
hard	O
choice	O
tasks	O
,	O
and	O
evaluate	O
how	O
this	O
effectiveness	O
changes	O
with	O
the	O
imposition	O
of	O
market	B-KEY
barriers	I-KEY
.	O

A	O
Reinforcement	B-KEY
Learning	I-KEY
based	O
Distributed	B-KEY
Search	I-KEY
Algorithm	I-KEY
For	O
Hierarchical	O
Peer-to-Peer	O
Information	O
Retrieval	O
Systems	O
ABSTRACT	O
The	O
dominant	O
existing	O
routing	O
strategies	O
employed	O
in	O
peerto-peer	O
-LRB-	O
P2P	O
-RRB-	O
based	O
information	O
retrieval	O
-LRB-	O
IR	O
-RRB-	O
systems	O
are	O
similarity-based	O
approaches	O
.	O
In	O
these	O
approaches	O
,	O
agents	O
depend	O
on	O
the	O
content	O
similarity	O
between	O
incoming	O
queries	B-KEY
and	O
their	O
direct	O
neighboring	O
agents	O
to	O
direct	O
the	O
distributed	O
search	O
sessions	O
.	O
However	O
,	O
such	O
a	O
heuristic	O
is	O
myopic	O
in	O
that	O
the	O
neighboring	O
agents	O
may	O
not	O
be	O
connected	O
to	O
more	O
relevant	O
agents	O
.	O
In	O
this	O
paper	O
,	O
an	O
online	O
reinforcement-learning	O
based	O
approach	O
is	O
developed	O
to	O
take	O
advantage	O
of	O
the	O
dynamic	O
run-time	O
characteristics	O
of	O
P2P	O
IR	O
systems	O
as	O
represented	O
by	O
information	O
about	O
past	O
search	O
sessions	O
.	O
Specifically	O
,	O
agents	O
maintain	O
estimates	O
on	O
the	O
downstream	O
agents	O
'	O
abilities	O
to	O
provide	O
relevant	O
documents	O
for	O
incoming	O
queries	B-KEY
.	O
These	O
estimates	O
are	O
updated	O
gradually	O
by	O
learning	O
from	O
the	O
feedback	O
information	O
returned	O
from	O
previous	O
search	O
sessions	O
.	O
Based	O
on	O
this	O
information	O
,	O
the	O
agents	O
derive	O
corresponding	O
routing	B-KEY
policies	I-KEY
.	O
Thereafter	O
,	O
these	O
agents	O
route	O
the	O
queries	B-KEY
based	O
on	O
the	O
learned	O
policies	O
and	O
update	O
the	O
estimates	O
based	O
on	O
the	O
new	O
routing	B-KEY
policies	I-KEY
.	O
Experimental	O
results	O
demonstrate	O
that	O
the	O
learning	B-KEY
algorithm	I-KEY
improves	O
considerably	O
the	O
routing	O
performance	O
on	O
two	O
test	O
collection	O
sets	O
that	O
have	O
been	O
used	O
in	O
a	O
variety	O
of	O
distributed	O
IR	O
studies	O
.	O

Computing	O
the	O
Optimal	B-KEY
Strategy	I-KEY
to	O
Commit	B-KEY
to	O
∗	O
ABSTRACT	O
In	O
multiagent	B-KEY
systems	I-KEY
,	O
strategic	O
settings	O
are	O
often	O
analyzed	O
under	O
the	O
assumption	O
that	O
the	O
players	O
choose	O
their	O
strategies	O
simultaneously	O
.	O
However	O
,	O
this	O
model	O
is	O
not	O
always	O
realistic	O
.	O
In	O
many	O
settings	O
,	O
one	O
player	O
is	O
able	O
to	O
commit	B-KEY
to	O
a	O
strategy	O
before	O
the	O
other	O
player	O
makes	O
a	O
decision	O
.	O
Such	O
models	O
are	O
synonymously	O
referred	O
to	O
as	O
leadership	B-KEY
,	O
commitment	B-KEY
,	O
or	O
Stackelberg	B-KEY
models	O
,	O
and	O
optimal	O
play	O
in	O
such	O
models	O
is	O
often	O
significantly	O
different	O
from	O
optimal	O
play	O
in	O
the	O
model	O
where	O
strategies	O
are	O
selected	O
simultaneously	O
.	O
The	O
recent	O
surge	O
in	O
interest	O
in	O
computing	O
game-theoretic	O
solutions	O
has	O
so	O
far	O
ignored	O
leadership	B-KEY
models	O
-LRB-	O
with	O
the	O
exception	O
of	O
the	O
interest	O
in	O
mechanism	O
design	O
,	O
where	O
the	O
designer	O
is	O
implicitly	O
in	O
a	O
leadership	O
position	O
-RRB-	O
.	O
In	O
this	O
paper	O
,	O
we	O
study	O
how	O
to	O
compute	O
optimal	B-KEY
strategies	I-KEY
to	O
commit	B-KEY
to	O
under	O
both	O
commitment	B-KEY
to	O
pure	B-KEY
strategies	I-KEY
and	O
commitment	B-KEY
to	O
mixed	B-KEY
strategies	I-KEY
,	O
in	O
both	O
normal-form	O
and	O
Bayesian	B-KEY
games	I-KEY
.	O
We	O
give	O
both	O
positive	O
results	O
-LRB-	O
efficient	O
algorithms	O
-RRB-	O
and	O
negative	O
results	O
-LRB-	O
NP-hardness	B-KEY
results	O
-RRB-	O
.	O

Edge	B-KEY
Indexing	I-KEY
in	O
a	O
Grid	O
for	O
Highly	O
Dynamic	B-KEY
Virtual	I-KEY
Environments	I-KEY
∗	O
ABSTRACT	O
Newly	O
emerging	O
game	O
--	O
based	O
application	O
systems	O
such	O
as	O
Second	O
Life1	O
provide	O
3D	O
virtual	O
environments	O
where	O
multiple	O
users	O
interact	O
with	O
each	O
other	O
in	O
real	O
--	O
time	O
.	O
They	O
are	O
filled	O
with	O
autonomous	O
,	O
mutable	B-KEY
virtual	I-KEY
content	I-KEY
which	O
is	O
continuously	O
augmented	O
by	O
the	O
users	O
.	O
To	O
make	O
the	O
systems	O
highly	O
scalable	O
and	O
dynamically	O
extensible	O
,	O
they	O
are	O
usually	O
built	O
on	O
a	O
client	O
--	O
server	O
based	O
grid	O
subspace	O
division	O
where	O
the	O
virtual	O
worlds	O
are	O
partitioned	O
into	O
manageable	O
sub	O
--	O
worlds	O
.	O
In	O
each	O
sub	O
--	O
world	O
,	O
the	O
user	O
continuously	O
receives	O
relevant	O
geometry	O
updates	O
of	O
moving	O
objects	O
from	O
remotely	O
connected	O
servers	O
and	O
renders	O
them	O
according	O
to	O
her	O
viewpoint	O
,	O
rather	O
than	O
retrieving	O
them	O
from	O
a	O
local	O
storage	O
medium	O
.	O
In	O
such	O
systems	O
,	O
the	O
determination	O
of	O
the	O
set	O
of	O
objects	O
that	O
are	O
visible	O
from	O
a	O
user	O
's	O
viewpoint	O
is	O
one	O
of	O
the	O
primary	O
factors	O
that	O
affect	O
server	O
throughput	O
and	O
scalability	O
.	O
Specifically	O
,	O
performing	O
real	O
--	O
time	O
visibility	O
tests	O
in	O
extremely	O
dynamic	B-KEY
virtual	I-KEY
environments	I-KEY
is	O
a	O
very	O
challenging	O
task	O
as	O
millions	O
of	O
objects	O
and	O
sub-millions	O
of	O
active	O
users	O
are	O
moving	O
and	O
interacting	O
.	O
We	O
recognize	O
that	O
the	O
described	O
challenges	O
are	O
closely	O
related	O
to	O
a	O
spatial	B-KEY
database	I-KEY
problem	O
,	O
and	O
hence	O
we	O
map	O
the	O
moving	O
geometry	O
objects	O
in	O
the	O
virtual	O
space	O
to	O
a	O
set	O
of	O
multi-dimensional	O
objects	O
in	O
a	O
spatial	B-KEY
database	I-KEY
while	O
modeling	O
each	O
avatar	O
both	O
as	O
a	O
spatial	O
object	O
and	O
a	O
moving	O
query	O
.	O
Unfortunately	O
,	O
existing	O
spatial	B-KEY
indexing	I-KEY
methods	O
are	O
unsuitable	O
for	O
this	O
kind	O
of	O
new	O
environments	O
.	O
The	O
main	O
goal	O
of	O
this	O
paper	O
is	O
to	O
present	O
an	O
efficient	O
spatial	B-KEY
index	I-KEY
structure	O
that	O
minimizes	O
unexpected	O
object	B-KEY
popping	I-KEY
and	O
supports	O
highly	O
scalable	O
real	O
--	O
time	O
visibility	O
determination	O
.	O
We	O
then	O
uncover	O
many	O
useful	O
properties	O
of	O
this	O
structure	O
and	O
compare	O
the	O
index	O
structure	O
with	O
various	O
spatial	B-KEY
indexing	I-KEY
methods	O
in	O
terms	O
of	O
query	O
quality	O
,	O
system	O
throughput	O
,	O
and	O
resource	O
utilization	O
.	O
We	O
expect	O
our	O
approach	O
to	O
lay	O
the	O
groundwork	O
for	O
next	O
--	O
generation	O
virtual	O
frameworks	O
that	O
may	O
merge	O
into	O
existing	O
web	O
--	O
based	O
services	O
in	O
the	O
near	O
future	O
.	O
∗	O
This	O
research	O
has	O
been	O
funded	O
in	O
part	O
by	O
NSF	O
grants	O
EEC9529152	O
-LRB-	O
IMSC	O
ERC	O
-RRB-	O
and	O
IIS-0534761	O
,	O
and	O
equipment	O
gifts	O
from	O
Intel	O
Corporation	O
,	O
Hewlett-Packard	O
,	O
Sun	O
Microsystems	O
and	O
Raptor	O
Networks	O
Technology	O
.	O
Categories	O
and	O
Subject	O
Descriptors	O
:	O
C.	O
2.4	O
-LSB-	O
Computer	O
--	O
Com	O

Combinatorial	B-KEY
Agency	I-KEY
ABSTRACT	O
Much	O
recent	O
research	O
concerns	O
systems	O
,	O
such	O
as	O
the	O
Internet	O
,	O
whose	O
components	O
are	O
owned	O
and	O
operated	O
by	O
different	O
parties	O
,	O
each	O
with	O
his	O
own	O
``	O
selfish	O
''	O
goal	O
.	O
The	O
field	O
of	O
Algorithmic	O
Mechanism	O
Design	O
handles	O
the	O
issue	O
of	O
private	O
information	O
held	O
by	O
the	O
different	O
parties	O
in	O
such	O
computational	O
settings	O
.	O
This	O
paper	O
deals	O
with	O
a	O
complementary	O
problem	O
in	O
such	O
settings	O
:	O
handling	O
the	O
``	O
hidden	O
actions	O
''	O
that	O
are	O
performed	O
by	O
the	O
different	O
parties	O
.	O
Our	O
model	O
is	O
a	O
combinatorial	O
variant	O
of	O
the	O
classical	O
principalagent	O
problem	O
from	O
economic	O
theory	O
.	O
In	O
our	O
setting	O
a	O
principal	O
must	O
motivate	O
a	O
team	O
of	O
strategic	O
agents	O
to	O
exert	O
costly	O
effort	O
on	O
his	O
behalf	O
,	O
but	O
their	O
actions	O
are	O
hidden	O
from	O
him	O
.	O
Our	O
focus	O
is	O
on	O
cases	O
where	O
complex	O
combinations	O
of	O
the	O
efforts	O
of	O
the	O
agents	O
influence	O
the	O
outcome	O
.	O
The	O
principal	O
motivates	O
the	O
agents	O
by	O
offering	O
to	O
them	O
a	O
set	O
of	O
contracts	O
,	O
which	O
together	O
put	O
the	O
agents	O
in	O
an	O
equilibrium	O
point	O
of	O
the	O
induced	O
game	O
.	O
We	O
present	O
formal	O
models	O
for	O
this	O
setting	O
,	O
suggest	O
and	O
embark	O
on	O
an	O
analysis	O
of	O
some	O
basic	O
issues	O
,	O
but	O
leave	O
many	O
questions	O
open	O
.	O

Strong	B-KEY
Equilibrium	I-KEY
in	O
Cost	B-KEY
Sharing	I-KEY
Connection	I-KEY
Games	I-KEY
*	O
ABSTRACT	O
In	O
this	O
work	O
we	O
study	O
cost	B-KEY
sharing	I-KEY
connection	I-KEY
games	I-KEY
,	O
where	O
each	O
player	O
has	O
a	O
source	O
and	O
sink	O
he	O
would	O
like	O
to	O
connect	O
,	O
and	O
the	O
cost	B-KEY
of	I-KEY
the	I-KEY
edges	I-KEY
is	O
either	O
shared	O
equally	O
-LRB-	O
fair	B-KEY
connection	I-KEY
games	I-KEY
-RRB-	O
or	O
in	O
an	O
arbitrary	O
way	O
-LRB-	O
general	B-KEY
connection	I-KEY
games	I-KEY
-RRB-	O
.	O
We	O
study	O
the	O
graph	B-KEY
topologies	I-KEY
that	O
guarantee	O
the	O
existence	O
of	O
a	O
strong	B-KEY
equilibrium	I-KEY
-LRB-	O
where	O
no	O
coalition	B-KEY
can	O
improve	O
the	O
cost	O
of	O
each	O
of	O
its	O
members	O
-RRB-	O
regardless	O
of	O
the	O
specific	B-KEY
costs	I-KEY
on	O
the	O
edges	O
.	O
Our	O
main	O
existence	O
results	O
are	O
the	O
following	O
:	O
-LRB-	O
1	O
-RRB-	O
For	O
a	O
single	B-KEY
source	I-KEY
and	I-KEY
sink	I-KEY
we	O
show	O
that	O
there	O
is	O
always	O
a	O
strong	B-KEY
equilibrium	I-KEY
-LRB-	O
both	O
for	O
fair	O
and	O
general	B-KEY
connection	I-KEY
games	I-KEY
-RRB-	O
.	O
-LRB-	O
2	O
-RRB-	O
For	O
a	O
single	B-KEY
source	I-KEY
multiple	I-KEY
sinks	I-KEY
we	O
show	O
that	O
for	O
a	O
series	O
parallel	O
graph	O
a	O
strong	B-KEY
equilibrium	I-KEY
always	O
exists	O
-LRB-	O
both	O
for	O
fair	O
and	O
general	B-KEY
connection	I-KEY
games	I-KEY
-RRB-	O
.	O
-LRB-	O
3	O
-RRB-	O
For	O
multi	B-KEY
source	I-KEY
and	I-KEY
sink	I-KEY
we	O
show	O
that	O
an	O
extension	B-KEY
parallel	I-KEY
graph	I-KEY
always	O
admits	O
a	O
strong	B-KEY
equilibrium	I-KEY
in	O
fair	B-KEY
connection	I-KEY
games	I-KEY
.	O
As	O
for	O
the	O
quality	O
of	O
the	O
strong	B-KEY
equilibrium	I-KEY
we	O
show	O
that	O
in	O
any	O
fair	B-KEY
connection	I-KEY
games	I-KEY
the	O
cost	O
of	O
a	O
strong	B-KEY
equilibrium	I-KEY
is	O
Θ	O
-LRB-	O
log	O
n	O
-RRB-	O
from	O
the	O
optimal	B-KEY
solution	I-KEY
,	O
where	O
n	O
is	O
the	O
number	B-KEY
of	I-KEY
players	I-KEY
.	O
-LRB-	O
This	O
should	O
be	O
contrasted	O
with	O
the	O
Ω	O
-LRB-	O
n	O
-RRB-	O
price	B-KEY
of	I-KEY
anarchy	I-KEY
for	O
the	O
same	O
setting	O
.	O
-RRB-	O
For	O
single	O
source	O
general	B-KEY
connection	I-KEY
games	I-KEY
and	O
single	O
source	O
single	O
sink	O
fair	B-KEY
connection	I-KEY
games	I-KEY
,	O
we	O
show	O
that	O
a	O
strong	B-KEY
equilibrium	I-KEY
is	O
always	O
an	O
optimal	B-KEY
solution	I-KEY
.	O
*	O
Research	O
supported	O
in	O
part	O
by	O
a	O
grant	O
of	O
the	O
Israel	O
Science	O
Foundation	O
,	O
Binational	O
Science	O
Foundation	O
-LRB-	O
BSF	O
-RRB-	O
,	O
GermanIsraeli	O
Foundation	O
-LRB-	O
GIF	O
-RRB-	O
,	O
Lady	O
Davis	O
Fellowship	O
,	O
an	O
IBM	O
faculty	O
award	O
,	O
and	O
the	O
IST	O
Programme	O
of	O
the	O
European	O
Community	O
,	O
under	O
the	O
PASCAL	O
Network	O
of	O
Excellence	O
,	O
IST-2002-506778	O
.	O
This	O
publication	O
only	O
reflects	O
the	O
authors	O
'	O
views	O
.	O

Commitment	B-KEY
and	O
Extortion	B-KEY
*	O
ABSTRACT	O
Making	O
commitments	B-KEY
,	O
e.g.	O
,	O
through	O
promises	O
and	O
threats	O
,	O
enables	O
a	O
player	O
to	O
exploit	O
the	O
strengths	O
of	O
his	O
own	O
strategic	B-KEY
position	I-KEY
as	O
well	O
as	O
the	O
weaknesses	O
of	O
that	O
of	O
his	O
opponents	O
.	O
Which	O
commitments	B-KEY
a	O
player	O
can	O
make	O
with	O
credibility	B-KEY
depends	O
on	O
the	O
circumstances	O
.	O
In	O
some	O
,	O
a	O
player	O
can	O
only	O
commit	B-KEY
to	O
the	O
performance	O
of	O
an	O
action	O
,	O
in	O
others	O
,	O
he	O
can	O
commit	B-KEY
himself	O
conditionally	O
on	O
the	O
actions	O
of	O
the	O
other	O
players	O
.	O
Some	O
situations	O
even	O
allow	O
for	O
commitments	B-KEY
on	O
commitments	B-KEY
or	O
for	O
commitments	B-KEY
to	O
randomized	O
actions	O
.	O
We	O
explore	O
the	O
formal	O
properties	O
of	O
these	O
types	O
of	O
-LRB-	O
conditional	O
-RRB-	O
commitment	B-KEY
and	O
their	O
interrelationships	O
.	O
So	O
as	O
to	O
preclude	O
inconsistencies	O
among	O
conditional	O
commitments	B-KEY
,	O
we	O
assume	O
an	O
order	O
in	O
which	O
the	O
players	O
make	O
their	O
commitments	B-KEY
.	O
Central	O
to	O
our	O
analyses	O
is	O
the	O
notion	O
of	O
an	O
extortion	B-KEY
,	O
which	O
we	O
define	O
,	O
for	O
a	O
given	O
order	O
of	O
the	O
players	O
,	O
as	O
a	O
profile	O
that	O
contains	O
,	O
for	O
each	O
player	O
,	O
an	O
optimal	O
commitment	B-KEY
given	O
the	O
commitments	B-KEY
of	O
the	O
players	O
that	O
committed	B-KEY
earlier	O
.	O
On	O
this	O
basis	O
,	O
we	O
investigate	O
for	O
different	O
commitment	B-KEY
types	O
whether	O
it	O
is	O
advantageous	O
to	O
commit	B-KEY
earlier	O
rather	O
than	O
later	O
,	O
and	O
how	O
the	O
outcomes	O
obtained	O
through	O
extortions	B-KEY
relate	O
to	O
backward	O
induction	O
and	O
Pareto	B-KEY
efficiency	I-KEY
.	O

Computing	O
the	O
Banzhaf	B-KEY
Power	I-KEY
Index	I-KEY
in	O
Network	O
Flow	O
Games	O
ABSTRACT	O
Preference	B-KEY
aggregation	I-KEY
is	O
used	O
in	O
a	O
variety	O
of	O
multiagent	B-KEY
applications	I-KEY
,	O
and	O
as	O
a	O
result	O
,	O
voting	B-KEY
theory	O
has	O
become	O
an	O
important	O
topic	O
in	O
multiagent	O
system	O
research	O
.	O
However	O
,	O
power	O
indices	O
-LRB-	O
which	O
reflect	O
how	O
much	O
``	O
real	O
power	O
''	O
a	O
voter	O
has	O
in	O
a	O
weighted	O
voting	B-KEY
system	O
-RRB-	O
have	O
received	O
relatively	O
little	O
attention	O
,	O
although	O
they	O
have	O
long	O
been	O
studied	O
in	O
political	O
science	O
and	O
economics	O
.	O
The	O
Banzhaf	B-KEY
power	I-KEY
index	I-KEY
is	O
one	O
of	O
the	O
most	O
popular	O
;	O
it	O
is	O
also	O
well-defined	O
for	O
any	O
simple	O
coalitional	O
game	O
.	O
In	O
this	O
paper	O
,	O
we	O
examine	O
the	O
computational	B-KEY
complexity	I-KEY
of	O
calculating	O
the	O
Banzhaf	B-KEY
power	I-KEY
index	I-KEY
within	O
a	O
particular	O
multiagent	O
domain	O
,	O
a	O
network	O
flow	O
game	O
.	O
Agents	O
control	O
the	O
edges	O
of	O
a	O
graph	O
;	O
a	O
coalition	O
wins	O
if	O
it	O
can	O
send	O
a	O
flow	O
of	O
a	O
given	O
size	O
from	O
a	O
source	O
vertex	O
to	O
a	O
target	O
vertex	O
.	O
The	O
relative	O
power	O
of	O
each	O
edge/agent	O
reflects	O
its	O
significance	O
in	O
enabling	O
such	O
a	O
flow	O
,	O
and	O
in	O
real-world	O
networks	O
could	O
be	O
used	O
,	O
for	O
example	O
,	O
to	O
allocate	O
resources	O
for	O
maintaining	O
parts	O
of	O
the	O
network	O
.	O
We	O
show	O
that	O
calculating	O
the	O
Banzhaf	B-KEY
power	I-KEY
index	I-KEY
of	O
each	O
agent	O
in	O
this	O
network	O
flow	O
domain	O
is	O
#P	O
-	O
complete	O
.	O
We	O
also	O
show	O
that	O
for	O
some	O
restricted	O
network	O
flow	O
domains	O
there	O
exists	O
a	O
polynomial	O
algorithm	O
to	O
calculate	O
agents	O
'	O
Banzhaf	O
power	O
indices	O
.	O

Reasoning	O
about	O
Judgment	O
and	O
Preference	B-KEY
Aggregation	I-KEY
◦	O
ABSTRACT	O
Agents	O
that	O
must	O
reach	O
agreements	O
with	O
other	O
agents	O
need	O
to	O
reason	O
about	O
how	O
their	O
preferences	O
,	O
judgments	O
,	O
and	O
beliefs	O
might	O
be	O
aggregated	O
with	O
those	O
of	O
others	O
by	O
the	O
social	O
choice	O
mechanisms	O
that	O
govern	O
their	O
interactions	O
.	O
The	O
recently	O
emerging	O
field	O
of	O
judgment	B-KEY
aggregation	I-KEY
studies	O
aggregation	O
from	O
a	O
logical	O
perspective	O
,	O
and	O
considers	O
how	O
multiple	O
sets	O
of	O
logical	O
formulae	O
can	O
be	O
aggregated	O
to	O
a	O
single	O
consistent	O
set	O
.	O
As	O
a	O
special	O
case	O
,	O
judgment	B-KEY
aggregation	I-KEY
can	O
be	O
seen	O
to	O
subsume	O
classical	O
preference	B-KEY
aggregation	I-KEY
.	O
We	O
present	O
a	O
modal	B-KEY
logic	I-KEY
that	O
is	O
intended	O
to	O
support	O
reasoning	O
about	O
judgment	B-KEY
aggregation	I-KEY
scenarios	O
-LRB-	O
and	O
hence	O
,	O
as	O
a	O
special	O
case	O
,	O
about	O
preference	B-KEY
aggregation	I-KEY
-RRB-	O
:	O
the	O
logical	O
language	O
is	O
interpreted	O
directly	O
in	O
judgment	B-KEY
aggregation	I-KEY
rules	O
.	O
We	O
present	O
a	O
sound	O
and	O
complete	B-KEY
axiomatisation	I-KEY
of	O
such	O
rules	O
.	O
We	O
show	O
that	O
the	O
logic	O
can	O
express	B-KEY
aggregation	O
rules	O
such	O
as	O
majority	O
voting	O
;	O
rule	O
properties	O
such	O
as	O
independence	O
;	O
and	O
results	O
such	O
as	O
the	O
discursive	B-KEY
paradox	I-KEY
,	O
Arrow	B-KEY
's	I-KEY
theorem	I-KEY
and	O
Condorcet	O
's	O
paradox	O
--	O
which	O
are	O
derivable	O
as	O
formal	O
theorems	O
of	O
the	O
logic	O
.	O
The	O
logic	O
is	O
parameterised	O
in	O
such	O
a	O
way	O
that	O
it	O
can	O
be	O
used	O
as	O
a	O
general	O
framework	O
for	O
comparing	O
the	O
logical	O
properties	O
of	O
different	O
types	O
of	O
aggregation	O
--	O
including	O
classical	O
preference	B-KEY
aggregation	I-KEY
.	O

Robust	O
Classification	O
of	O
Rare	O
Queries	O
Using	O
Web	O
Knowledge	O
ABSTRACT	O
We	O
propose	O
a	O
methodology	O
for	O
building	O
a	O
practical	O
robust	O
query	B-KEY
classification	I-KEY
system	O
that	O
can	O
identify	O
thousands	O
of	O
query	O
classes	O
with	O
reasonable	O
accuracy	O
,	O
while	O
dealing	O
in	O
realtime	O
with	O
the	O
query	O
volume	O
of	O
a	O
commercial	O
web	B-KEY
search	I-KEY
engine	O
.	O
We	O
use	O
a	O
blind	O
feedback	O
technique	O
:	O
given	O
a	O
query	O
,	O
we	O
determine	O
its	O
topic	O
by	O
classifying	O
the	O
web	B-KEY
search	I-KEY
results	O
retrieved	O
by	O
the	O
query	O
.	O
Motivated	O
by	O
the	O
needs	O
of	O
search	B-KEY
advertising	I-KEY
,	O
we	O
primarily	O
focus	O
on	O
rare	O
queries	O
,	O
which	O
are	O
the	O
hardest	O
from	O
the	O
point	O
of	O
view	O
of	O
machine	B-KEY
learning	I-KEY
,	O
yet	O
in	O
aggregation	O
account	O
for	O
a	O
considerable	O
fraction	O
of	O
search	B-KEY
engine	I-KEY
traffic	O
.	O
Empirical	O
evaluation	O
confirms	O
that	O
our	O
methodology	O
yields	O
a	O
considerably	O
higher	O
classification	O
accuracy	O
than	O
previously	O
reported	O
.	O
We	O
believe	O
that	O
the	O
proposed	O
methodology	O
will	O
lead	O
to	O
better	O
matching	O
of	O
online	O
ads	O
to	O
rare	O
queries	O
and	O
overall	O
to	O
a	O
better	O
user	O
experience	O
.	O

Collaboration	O
Among	O
a	O
Satellite	B-KEY
Swarm	I-KEY
ABSTRACT	O
The	O
paper	O
deals	O
with	O
on-board	B-KEY
planning	I-KEY
for	O
a	O
satellite	B-KEY
swarm	I-KEY
via	O
communication	B-KEY
and	I-KEY
negotiation	I-KEY
.	O
We	O
aim	O
at	O
defining	O
individual	O
behaviours	O
that	O
result	O
in	O
a	O
global	O
behaviour	O
that	O
meets	O
the	O
mission	O
requirements	O
.	O
We	O
will	O
present	O
the	O
formalization	O
of	O
the	O
problem	O
,	O
a	O
communication	O
protocol	O
,	O
a	O
solving	O
method	O
based	O
on	O
reactive	B-KEY
decision	I-KEY
rules	I-KEY
,	O
and	O
first	O
results	O
.	O

An	O
Advanced	B-KEY
Bidding	I-KEY
Agent	I-KEY
for	O
Advertisement	O
Selection	O
on	O
Public	O
Displays	O
ABSTRACT	O
In	O
this	O
paper	O
we	O
present	O
an	O
advanced	B-KEY
bidding	I-KEY
agent	I-KEY
that	O
participates	O
in	O
first-price	O
sealed	O
bid	O
auctions	O
to	O
allocate	O
advertising	O
space	O
on	O
BluScreen	O
--	O
an	O
experimental	O
public	O
advertisement	O
system	O
that	O
detects	O
users	O
through	O
the	O
presence	O
of	O
their	O
Bluetooth	O
enabled	O
devices	O
.	O
Our	O
bidding	B-KEY
agent	I-KEY
is	O
able	O
to	O
build	O
probabilistic	B-KEY
models	I-KEY
of	O
both	O
the	O
behaviour	O
of	O
users	O
who	O
view	O
the	O
adverts	O
,	O
and	O
the	O
auctions	B-KEY
that	O
it	O
participates	O
within	O
.	O
It	O
then	O
uses	O
these	O
models	O
to	O
maximise	O
the	O
exposure	O
that	O
its	O
adverts	O
receive	O
.	O
We	O
evaluate	O
the	O
effectiveness	O
of	O
this	O
bidding	B-KEY
agent	I-KEY
through	O
simulation	O
against	O
a	O
range	O
of	O
alternative	O
selection	O
mechanisms	O
including	O
a	O
simple	O
bidding	O
strategy	O
,	O
random	O
allocation	O
,	O
and	O
a	O
centralised	B-KEY
optimal	I-KEY
allocation	I-KEY
with	O
perfect	O
foresight	O
.	O
Our	O
bidding	B-KEY
agent	I-KEY
significantly	O
outperforms	O
both	O
the	O
simple	O
bidding	O
strategy	O
and	O
the	O
random	O
allocation	O
,	O
and	O
in	O
a	O
mixed	O
population	O
of	O
agents	O
it	O
is	O
able	O
to	O
expose	O
its	O
adverts	O
to	O
25	O
%	O
more	O
users	O
than	O
the	O
simple	O
bidding	O
strategy	O
.	O
Moreover	O
,	O
its	O
performance	O
is	O
within	O
7.5	O
%	O
of	O
that	O
of	O
the	O
centralised	B-KEY
optimal	I-KEY
allocation	I-KEY
despite	O
the	O
highly	O
uncertain	O
environment	O
in	O
which	O
it	O
must	O
operate	O
.	O

The	O
Impact	O
of	O
Caching	B-KEY
on	O
Search	O
Engines	O
ABSTRACT	O
In	O
this	O
paper	O
we	O
study	O
the	O
trade-offs	O
in	O
designing	O
efficient	O
caching	B-KEY
systems	O
for	O
Web	O
search	O
engines	O
.	O
We	O
explore	O
the	O
impact	O
of	O
different	O
approaches	O
,	O
such	O
as	O
static	O
vs.	O
dynamic	B-KEY
caching	I-KEY
,	O
and	O
caching	O
query	O
results	O
vs.	O
caching	O
posting	O
lists	O
.	O
Using	O
a	O
query	B-KEY
log	I-KEY
spanning	O
a	O
whole	O
year	O
we	O
explore	O
the	O
limitations	O
of	O
caching	B-KEY
and	O
we	O
demonstrate	O
that	O
caching	B-KEY
posting	O
lists	O
can	O
achieve	O
higher	O
hit	O
rates	O
than	O
caching	O
query	O
answers	O
.	O
We	O
propose	O
a	O
new	O
algorithm	O
for	O
static	B-KEY
caching	I-KEY
of	O
posting	O
lists	O
,	O
which	O
outperforms	O
previous	O
methods	O
.	O
We	O
also	O
study	O
the	O
problem	O
of	O
finding	O
the	O
optimal	O
way	O
to	O
split	O
the	O
static	B-KEY
cache	I-KEY
between	O
answers	O
and	O
posting	O
lists	O
.	O
Finally	O
,	O
we	O
measure	O
how	O
the	O
changes	O
in	O
the	O
query	B-KEY
log	I-KEY
affect	O
the	O
effectiveness	B-KEY
of	I-KEY
static	I-KEY
caching	I-KEY
,	O
given	O
our	O
observation	O
that	O
the	O
distribution	O
of	O
the	O
queries	O
changes	O
slowly	O
over	O
time	O
.	O
Our	O
results	O
and	O
observations	O
are	O
applicable	O
to	O
different	O
levels	O
of	O
the	O
data-access	B-KEY
hierarchy	I-KEY
,	O
for	O
instance	O
,	O
for	O
a	O
memory/disk	O
layer	O
or	O
a	O
broker/remote	O
server	O
layer	O
.	O

Revenue	B-KEY
Analysis	O
of	O
a	O
Family	O
of	O
Ranking	B-KEY
Rules	I-KEY
for	O
Keyword	B-KEY
Auctions	I-KEY
ABSTRACT	O
Keyword	B-KEY
auctions	I-KEY
lie	O
at	O
the	O
core	O
of	O
the	O
business	O
models	O
of	O
today	O
's	O
leading	O
search	B-KEY
engines	I-KEY
.	O
Advertisers	B-KEY
bid	O
for	O
placement	O
alongside	O
search	O
results	O
,	O
and	O
are	O
charged	O
for	O
clicks	O
on	O
their	O
ads	O
.	O
Advertisers	B-KEY
are	O
typically	O
ranked	O
according	O
to	O
a	O
score	O
that	O
takes	O
into	O
account	O
their	O
bids	O
and	O
potential	O
clickthrough	O
rates	O
.	O
We	O
consider	O
a	O
family	O
of	O
ranking	B-KEY
rules	I-KEY
that	O
contains	O
those	O
typically	O
used	O
to	O
model	O
Yahoo!	O
and	O
Google	O
's	O
auction	O
designs	O
as	O
special	O
cases	O
.	O
We	O
find	O
that	O
in	O
general	O
neither	O
of	O
these	O
is	O
necessarily	O
revenue-optimal	O
in	O
equilibrium	O
,	O
and	O
that	O
the	O
choice	O
of	O
ranking	B-KEY
rule	I-KEY
can	O
be	O
guided	O
by	O
considering	O
the	O
correlation	O
between	O
bidders	O
'	O
values	O
and	O
click-through	O
rates	O
.	O
We	O
propose	O
a	O
simple	O
approach	O
to	O
determine	O
a	O
revenue-optimal	O
ranking	B-KEY
rule	I-KEY
within	O
our	O
family	O
,	O
taking	O
into	O
account	O
effects	O
on	O
advertiser	B-KEY
satisfaction	O
and	O
user	O
experience	O
.	O
We	O
illustrate	O
the	O
approach	O
using	O
Monte-Carlo	O
simulations	O
based	O
on	O
distributions	O
fitted	O
to	O
Yahoo!	O
bid	O
and	O
click-through	O
rate	O
data	O
for	O
a	O
high-volume	O
keyword	O
.	O

Generalized	O
Value	O
Decomposition	O
and	O
Structured	O
Multiattribute	B-KEY
Auctions	I-KEY
ABSTRACT	O
Multiattribute	B-KEY
auction	I-KEY
mechanisms	O
generally	O
either	O
remain	O
agnostic	O
about	O
traders	O
'	O
preferences	O
,	O
or	O
presume	O
highly	O
restrictive	O
forms	O
,	O
such	O
as	O
full	O
additivity	O
.	O
Real	O
preferences	O
often	O
exhibit	O
dependencies	O
among	O
attributes	O
,	O
yet	O
may	O
possess	O
some	O
structure	O
that	O
can	O
be	O
usefully	O
exploited	O
to	O
streamline	O
communication	O
and	O
simplify	O
operation	O
of	O
a	O
multiattribute	B-KEY
auction	I-KEY
.	O
We	O
develop	O
such	O
a	O
structure	O
using	O
the	O
theory	B-KEY
of	I-KEY
measurable	I-KEY
value	I-KEY
functions	I-KEY
,	O
a	O
cardinal	O
utility	O
representation	O
based	O
on	O
an	O
underlying	O
order	O
over	O
preference	O
differences	O
.	O
A	O
set	O
of	O
local	O
conditional	O
independence	O
relations	O
over	O
such	O
differences	O
supports	O
a	O
generalized	O
additive	O
preference	O
representation	O
,	O
which	O
decomposes	O
utility	O
across	O
overlapping	O
clusters	O
of	O
related	O
attributes	O
.	O
We	O
introduce	O
an	O
iterative	O
auction	B-KEY
mechanism	O
that	O
maintains	O
prices	O
on	O
local	O
clusters	O
of	O
attributes	O
rather	O
than	O
the	O
full	O
space	O
of	O
joint	O
configurations	O
.	O
When	O
traders	O
'	O
preferences	O
are	O
consistent	O
with	O
the	O
auction	B-KEY
's	O
generalized	O
additive	O
structure	O
,	O
the	O
mechanism	O
produces	O
approximately	O
optimal	O
allocations	O
,	O
at	O
approximate	O
VCG	O
prices	O
.	O

Bidding	O
Optimally	O
in	O
Concurrent	O
Second-Price	O
Auctions	O
of	O
Perfectly	O
Substitutable	O
Goods	O
ABSTRACT	O
We	O
derive	O
optimal	B-KEY
bidding	I-KEY
strategies	I-KEY
for	O
a	O
global	B-KEY
bidding	I-KEY
agent	I-KEY
that	O
participates	O
in	O
multiple	O
,	O
simultaneous	O
second-price	O
auctions	O
with	O
perfect	B-KEY
substitutes	I-KEY
.	O
We	O
first	O
consider	O
a	O
model	O
where	O
all	O
other	O
bidders	O
are	O
local	O
and	O
participate	O
in	O
a	O
single	O
auction	O
.	O
For	O
this	O
case	O
,	O
we	O
prove	O
that	O
,	O
assuming	O
free	O
disposal	O
,	O
the	O
global	O
bidder	O
should	O
always	O
place	O
non-zero	O
bids	O
in	O
all	O
available	O
auctions	O
,	O
irrespective	O
of	O
the	O
local	O
bidders	O
'	O
valuation	O
distribution	O
.	O
Furthermore	O
,	O
for	O
non-decreasing	B-KEY
valuation	I-KEY
distributions	I-KEY
,	O
we	O
prove	O
that	O
the	O
problem	O
of	O
finding	O
the	O
optimal	O
bids	O
reduces	O
to	O
two	O
dimensions	O
.	O
These	O
results	O
hold	O
both	O
in	O
the	O
case	O
where	O
the	O
number	O
of	O
local	O
bidders	O
is	O
known	O
and	O
when	O
this	O
number	O
is	O
determined	O
by	O
a	O
Poisson	O
distribution	O
.	O
This	O
analysis	O
extends	O
to	O
online	B-KEY
markets	I-KEY
where	O
,	O
typically	O
,	O
auctions	O
occur	O
both	O
concurrently	O
and	O
sequentially	O
.	O
In	O
addition	O
,	O
by	O
combining	O
analytical	O
and	O
simulation	O
results	O
,	O
we	O
demonstrate	O
that	O
similar	O
results	O
hold	O
in	O
the	O
case	O
of	O
several	O
global	O
bidders	O
,	O
provided	O
that	O
the	O
market	O
consists	O
of	O
both	O
global	O
and	O
local	O
bidders	O
.	O
Finally	O
,	O
we	O
address	O
the	O
efficiency	O
of	O
the	O
overall	O
market	O
,	O
and	O
show	O
that	O
information	O
about	O
the	O
number	O
of	O
local	O
bidders	O
is	O
an	O
important	O
determinant	O
for	O
the	O
way	O
in	O
which	O
a	O
global	O
bidder	O
affects	O
efficiency	O
.	O

Learning	B-KEY
From	I-KEY
Revealed	I-KEY
Preference	I-KEY
ABSTRACT	O
A	O
sequence	O
of	O
prices	O
and	O
demands	O
are	O
rationalizable	B-KEY
if	O
there	O
exists	O
a	O
concave	O
,	O
continuous	O
and	O
monotone	O
utility	O
function	O
such	O
that	O
the	O
demands	O
are	O
the	O
maximizers	O
of	O
the	O
utility	O
function	O
over	O
the	O
budget	O
set	O
corresponding	O
to	O
the	O
price	O
.	O
Afriat	O
-LSB-	O
1	O
-RSB-	O
presented	O
necessary	O
and	O
sufficient	O
conditions	O
for	O
a	O
finite	O
sequence	O
to	O
be	O
rationalizable	B-KEY
.	O
Varian	O
-LSB-	O
20	O
-RSB-	O
and	O
later	O
Blundell	O
et	O
al.	O
-LSB-	O
3	O
,	O
4	O
-RSB-	O
continued	O
this	O
line	O
of	O
work	O
studying	O
nonparametric	O
methods	O
to	O
forecasts	B-KEY
demand	O
.	O
Their	O
results	O
essentially	O
characterize	O
learnability	O
of	O
degenerate	O
classes	O
of	O
demand	B-KEY
functions	I-KEY
and	O
therefore	O
fall	O
short	O
of	O
giving	O
a	O
general	O
degree	O
of	O
confidence	O
in	O
the	O
forecast	B-KEY
.	O
The	O
present	O
paper	O
complements	O
this	O
line	O
of	O
research	O
by	O
introducing	O
a	O
statistical	O
model	O
and	O
a	O
measure	O
of	O
complexity	O
through	O
which	O
we	O
are	O
able	O
to	O
study	O
the	O
learnability	O
of	O
classes	O
of	O
demand	B-KEY
functions	I-KEY
and	O
derive	O
a	O
degree	O
of	O
confidence	O
in	O
the	O
forecasts	B-KEY
.	O
Our	O
results	O
show	O
that	O
the	O
class	O
of	O
all	O
demand	B-KEY
functions	I-KEY
has	O
unbounded	O
complexity	O
and	O
therefore	O
is	O
not	O
learnable	O
,	O
but	O
that	O
there	O
exist	O
interesting	O
and	O
potentially	O
useful	O
classes	O
that	O
are	O
learnable	O
from	O
finite	O
samples	O
.	O
We	O
also	O
present	O
a	O
learning	O
algorithm	O
that	O
is	O
an	O
adaptation	O
of	O
a	O
new	O
proof	O
of	O
Afriat	O
's	O
theorem	O
due	O
to	O
Teo	O
and	O
Vohra	O
-LSB-	O
17	O
-RSB-	O
.	O

Clearing	O
Algorithms	O
for	O
Barter	B-KEY
Exchange	B-KEY
Markets	O
:	O
Enabling	O
Nationwide	O
Kidney	O
Exchanges	B-KEY
ABSTRACT	O
In	O
barter-exchange	O
markets	O
,	O
agents	O
seek	O
to	O
swap	O
their	O
items	O
with	O
one	O
another	O
,	O
in	O
order	O
to	O
improve	O
their	O
own	O
utilities	O
.	O
These	O
swaps	O
consist	O
of	O
cycles	O
of	O
agents	O
,	O
with	O
each	O
agent	O
receiving	O
the	O
item	O
of	O
the	O
next	O
agent	O
in	O
the	O
cycle	O
.	O
We	O
focus	O
mainly	O
on	O
the	O
upcoming	O
national	O
kidney-exchange	O
market	O
,	O
where	O
patients	O
with	O
kidney	O
disease	O
can	O
obtain	O
compatible	O
donors	O
by	O
swapping	O
their	O
own	O
willing	O
but	O
incompatible	O
donors	O
.	O
With	O
over	O
70,000	O
patients	O
already	O
waiting	O
for	O
a	O
cadaver	O
kidney	O
in	O
the	O
US	O
,	O
this	O
market	O
is	O
seen	O
as	O
the	O
only	O
ethical	O
way	O
to	O
significantly	O
reduce	O
the	O
4,000	O
deaths	O
per	O
year	O
attributed	O
to	O
kidney	O
disease	O
.	O
The	O
clearing	O
problem	O
involves	O
finding	O
a	O
social	O
welfare	O
maximizing	O
exchange	B-KEY
when	O
the	O
maximum	O
length	O
of	O
a	O
cycle	O
is	O
fixed	O
.	O
Long	O
cycles	O
are	O
forbidden	O
,	O
since	O
,	O
for	O
incentive	O
reasons	O
,	O
all	O
transplants	B-KEY
in	O
a	O
cycle	O
must	O
be	O
performed	O
simultaneously	O
.	O
Also	O
,	O
in	O
barter-exchanges	O
generally	O
,	O
more	O
agents	O
are	O
affected	O
if	O
one	O
drops	O
out	O
of	O
a	O
longer	O
cycle	O
.	O
We	O
prove	O
that	O
the	O
clearing	O
problem	O
with	O
this	O
cycle-length	O
constraint	O
is	O
NP-hard	O
.	O
Solving	O
it	O
exactly	O
is	O
one	O
of	O
the	O
main	O
challenges	O
in	O
establishing	O
a	O
national	O
kidney	O
exchange	B-KEY
.	O
We	O
present	O
the	O
first	O
algorithm	O
capable	O
of	O
clearing	O
these	O
markets	O
on	O
a	O
nationwide	O
scale	O
.	O
The	O
key	O
is	O
incremental	O
problem	O
formulation	O
.	O
We	O
adapt	O
two	O
paradigms	O
for	O
the	O
task	O
:	O
constraint	O
generation	O
and	O
column	B-KEY
generation	I-KEY
.	O
For	O
each	O
,	O
we	O
develop	O
techniques	O
that	O
dramatically	O
improve	O
both	O
runtime	O
and	O
memory	O
usage	O
.	O
We	O
conclude	O
that	O
column	B-KEY
generation	I-KEY
scales	O
drastically	O
better	O
than	O
constraint	O
generation	O
.	O
Our	O
algorithm	O
also	O
supports	O
several	O
generalizations	O
,	O
as	O
demanded	O
by	O
real-world	O
kidney	O
exchanges	B-KEY
.	O
Our	O
algorithm	O
replaced	O
CPLEX	O
as	O
the	O
clearing	O
algorithm	O
of	O
the	O
Alliance	O
for	O
Paired	O
Donation	O
,	O
one	O
of	O
the	O
leading	O
kidney	O
exchanges	B-KEY
.	O
The	O
match	B-KEY
runs	O
are	O
conducted	O
every	O
two	O
weeks	O
and	O
transplants	B-KEY
based	O
on	O
our	O
optimizations	O
have	O
already	O
been	O
conducted	O
.	O

Learn	O
from	O
Web	O
Search	O
Logs	O
to	O
Organize	O
Search	O
Results	O
ABSTRACT	O
Effective	O
organization	O
of	O
search	O
results	O
is	O
critical	O
for	O
improving	O
the	O
utility	O
of	O
any	O
search	O
engine	O
.	O
Clustering	O
search	O
results	O
is	O
an	O
effective	O
way	O
to	O
organize	O
search	O
results	O
,	O
which	O
allows	O
a	O
user	O
to	O
navigate	O
into	O
relevant	O
documents	O
quickly	O
.	O
However	O
,	O
two	O
deficiencies	O
of	O
this	O
approach	O
make	O
it	O
not	O
always	O
work	O
well	O
:	O
-LRB-	O
1	O
-RRB-	O
the	O
clusters	O
discovered	O
do	O
not	O
necessarily	O
correspond	O
to	O
the	O
interesting	B-KEY
aspects	I-KEY
of	O
a	O
topic	O
from	O
the	O
user	O
's	O
perspective	O
;	O
and	O
-LRB-	O
2	O
-RRB-	O
the	O
cluster	O
labels	O
generated	O
are	O
not	O
informative	O
enough	O
to	O
allow	O
a	O
user	O
to	O
identify	O
the	O
right	O
cluster	O
.	O
In	O
this	O
paper	O
,	O
we	O
propose	O
to	O
address	O
these	O
two	O
deficiencies	O
by	O
-LRB-	O
1	O
-RRB-	O
learning	O
``	O
interesting	B-KEY
aspects	I-KEY
''	O
of	O
a	O
topic	O
from	O
Web	O
search	O
logs	O
and	O
organizing	O
search	O
results	O
accordingly	O
;	O
and	O
-LRB-	O
2	O
-RRB-	O
generating	O
more	O
meaningful	B-KEY
cluster	I-KEY
labels	I-KEY
using	O
past	B-KEY
query	I-KEY
words	O
entered	O
by	O
users	O
.	O
We	O
evaluate	O
our	O
proposed	O
method	O
on	O
a	O
commercial	O
search	B-KEY
engine	I-KEY
log	I-KEY
data	O
.	O
Compared	O
with	O
the	O
traditional	O
methods	O
of	O
clustering	O
search	O
results	O
,	O
our	O
method	O
can	O
give	O
better	O
result	O
organization	O
and	O
more	O
meaningful	O
labels	O
.	O

New	B-KEY
Event	I-KEY
Detection	I-KEY
Based	O
on	O
Indexing-tree	O
and	O
Named	B-KEY
Entity	I-KEY
ABSTRACT	O
New	B-KEY
Event	I-KEY
Detection	I-KEY
-LRB-	O
NED	O
-RRB-	O
aims	O
at	O
detecting	O
from	O
one	O
or	O
multiple	O
streams	O
of	O
news	O
stories	O
that	O
which	O
one	O
is	O
reported	O
on	O
a	O
new	O
event	O
-LRB-	O
i.e.	O
not	O
reported	O
previously	O
-RRB-	O
.	O
With	O
the	O
overwhelming	O
volume	O
of	O
news	O
available	O
today	O
,	O
there	O
is	O
an	O
increasing	O
need	O
for	O
a	O
NED	O
system	O
which	O
is	O
able	O
to	O
detect	O
new	O
events	O
more	O
efficiently	O
and	O
accurately	O
.	O
In	O
this	O
paper	O
we	O
propose	O
a	O
new	O
NED	O
model	O
to	O
speed	O
up	O
the	O
NED	O
task	O
by	O
using	O
news	O
indexing-tree	O
dynamically	O
.	O
Moreover	O
,	O
based	O
on	O
the	O
observation	O
that	O
terms	O
of	O
different	O
types	O
have	O
different	O
effects	O
for	O
NED	O
task	O
,	O
two	O
term	B-KEY
reweighting	I-KEY
approaches	I-KEY
are	O
proposed	O
to	O
improve	O
NED	B-KEY
accuracy	I-KEY
.	O
In	O
the	O
first	O
approach	O
,	O
we	O
propose	O
to	O
adjust	O
term	B-KEY
weights	I-KEY
dynamically	O
based	O
on	O
previous	O
story	O
clusters	O
and	O
in	O
the	O
second	O
approach	O
,	O
we	O
propose	O
to	O
employ	O
statistics	B-KEY
on	O
training	B-KEY
data	I-KEY
to	O
learn	O
the	O
named	B-KEY
entity	I-KEY
reweighting	O
model	O
for	O
each	O
class	B-KEY
of	I-KEY
stories	I-KEY
.	O
Experimental	O
results	O
on	O
two	O
Linguistic	B-KEY
Data	I-KEY
Consortium	I-KEY
-LRB-	O
LDC	O
-RRB-	O
datasets	O
TDT2	O
and	O
TDT3	O
show	O
that	O
the	O
proposed	O
model	O
can	O
improve	O
both	O
efficiency	O
and	O
accuracy	O
of	O
NED	O
task	O
significantly	O
,	O
compared	O
to	O
the	O
baseline	B-KEY
system	I-KEY
and	O
other	O
existing	B-KEY
systems	I-KEY
.	O

Resolving	O
Conflict	O
and	O
Inconsistency	O
in	O
Norm-Regulated	O
Virtual	B-KEY
Organizations	I-KEY
ABSTRACT	O
Norm-governed	O
virtual	B-KEY
organizations	I-KEY
define	O
,	O
govern	O
and	O
facilitate	O
coordinated	O
resource	O
sharing	O
and	O
problem	O
solving	O
in	O
societies	O
of	O
agents	B-KEY
.	O
With	O
an	O
explicit	O
account	O
of	O
norms	O
,	O
openness	O
in	O
virtual	B-KEY
organizations	I-KEY
can	O
be	O
achieved	O
:	O
new	O
components	O
,	O
designed	O
by	O
various	O
parties	O
,	O
can	O
be	O
seamlessly	O
accommodated	O
.	O
We	O
focus	O
on	O
virtual	B-KEY
organizations	I-KEY
realised	O
as	O
multi-agent	O
systems	O
,	O
in	O
which	O
human	O
and	O
software	O
agents	B-KEY
interact	O
to	O
achieve	O
individual	O
and	O
global	O
goals	O
.	O
However	O
,	O
any	O
realistic	O
account	O
of	O
norms	O
should	O
address	O
their	O
dynamic	O
nature	O
:	O
norms	O
will	O
change	O
as	O
agents	B-KEY
interact	O
with	O
each	O
other	O
and	O
their	O
environment	O
.	O
Due	O
to	O
the	O
changing	O
nature	O
of	O
norms	O
or	O
due	O
to	O
norms	O
stemming	O
from	O
different	O
virtual	B-KEY
organizations	I-KEY
,	O
there	O
will	O
be	O
situations	O
when	O
an	O
action	O
is	O
simultaneously	O
permitted	O
and	O
prohibited	O
,	O
that	O
is	O
,	O
a	O
conflict	O
arises	O
.	O
Likewise	O
,	O
there	O
will	O
be	O
situations	O
when	O
an	O
action	O
is	O
both	O
obliged	O
and	O
prohibited	O
,	O
that	O
is	O
,	O
an	O
inconsistency	O
arises	O
.	O
We	O
introduce	O
an	O
approach	O
,	O
based	O
on	O
first-order	O
unification	O
,	O
to	O
detect	O
and	O
resolve	O
such	O
conflicts	O
and	O
inconsistencies	O
.	O
In	O
our	O
proposed	O
solution	O
,	O
we	O
annotate	O
a	O
norm	O
with	O
the	O
set	O
of	O
values	O
their	O
variables	O
should	O
not	O
have	O
in	O
order	O
to	O
avoid	O
a	O
conflict	O
or	O
an	O
inconsistency	O
with	O
another	O
norm	O
.	O
Our	O
approach	O
neatly	O
accommodates	O
the	O
domain-dependent	O
interrelations	O
among	O
actions	O
and	O
the	O
indirect	O
conflicts/inconsistencies	O
these	O
may	O
cause	O
.	O
More	O
generally	O
,	O
we	O
can	O
capture	O
a	O
useful	O
notion	O
of	O
inter-agent	O
-LRB-	O
and	O
inter-role	O
-RRB-	O
delegation	O
of	O
actions	O
and	O
norms	O
associated	O
to	O
them	O
,	O
and	O
use	O
it	O
to	O
address	O
conflicts/inconsistencies	O
caused	O
by	O
action	O
delegation	O
.	O
We	O
illustrate	O
our	O
approach	O
with	O
an	O
e-Science	O
example	O
in	O
which	O
agents	B-KEY
support	O
Grid	O
services	O
.	O

Regularized	B-KEY
Clustering	O
for	O
Documents	O
*	O
ABSTRACT	O
In	O
recent	O
years	O
,	O
document	B-KEY
clustering	I-KEY
has	O
been	O
receiving	O
more	O
and	O
more	O
attentions	O
as	O
an	O
important	O
and	O
fundamental	O
technique	O
for	O
unsupervised	O
document	O
organization	O
,	O
automatic	O
topic	O
extraction	O
,	O
and	O
fast	O
information	O
retrieval	O
or	O
filtering	O
.	O
In	O
this	O
paper	O
,	O
we	O
propose	O
a	O
novel	O
method	O
for	O
clustering	O
documents	O
using	O
regularization	B-KEY
.	O
Unlike	O
traditional	O
globally	B-KEY
regularized	I-KEY
clustering	O
methods	O
,	O
our	O
method	O
first	O
construct	O
a	O
local	O
regularized	O
linear	O
label	O
predictor	O
for	O
each	O
document	O
vector	O
,	O
and	O
then	O
combine	O
all	O
those	O
local	O
regularizers	O
with	O
a	O
global	O
smoothness	O
regularizer	O
.	O
So	O
we	O
call	O
our	O
algorithm	O
Clustering	O
with	O
Local	O
and	O
Global	B-KEY
Regularization	I-KEY
-LRB-	O
CLGR	O
-RRB-	O
.	O
We	O
will	O
show	O
that	O
the	O
cluster	O
memberships	O
of	O
the	O
documents	O
can	O
be	O
achieved	O
by	O
eigenvalue	O
decomposition	O
of	O
a	O
sparse	O
symmetric	O
matrix	O
,	O
which	O
can	O
be	O
efficiently	O
solved	O
by	O
iterative	O
methods	O
.	O
Finally	O
our	O
experimental	O
evaluations	O
on	O
several	O
datasets	O
are	O
presented	O
to	O
show	O
the	O
superiorities	O
of	O
CLGR	O
over	O
traditional	O
document	B-KEY
clustering	I-KEY
methods	O
.	O

Implementation	O
and	O
Performance	O
Evaluation	O
of	O
CONFLEX-G	B-KEY
:	O
Grid-enabled	O
Molecular	O
Conformational	B-KEY
Space	I-KEY
Search	I-KEY
Program	O
with	O
OmniRPC	B-KEY
ABSTRACT	O
CONFLEX-G	B-KEY
is	O
the	O
grid-enabled	O
version	O
of	O
a	O
molecular	O
conformational	B-KEY
space	I-KEY
search	I-KEY
program	O
called	O
CONFLEX	O
.	O
We	O
have	O
implemented	O
CONFLEX-G	B-KEY
using	O
a	O
grid	B-KEY
RPC	I-KEY
system	I-KEY
called	O
OmniRPC	B-KEY
.	O
In	O
this	O
paper	O
,	O
we	O
report	O
the	O
performance	O
of	O
CONFLEX-G	B-KEY
in	O
a	O
grid	O
testbed	O
of	O
several	O
geographically	O
distributed	O
PC	B-KEY
clusters	I-KEY
.	O
In	O
order	O
to	O
explore	O
many	O
conformation	O
of	O
large	O
bio-molecules	B-KEY
,	O
CONFLEX-G	B-KEY
generates	O
trial	O
structures	O
of	O
the	O
molecules	O
and	O
allocates	O
jobs	O
to	O
optimize	O
a	O
trial	O
structure	O
with	O
a	O
reliable	O
molecular	B-KEY
mechanics	I-KEY
method	O
in	O
the	O
grid	O
.	O
OmniRPC	B-KEY
provides	O
a	O
restricted	O
persistence	O
model	O
to	O
support	O
the	O
parametric	O
search	O
applications	O
.	O
In	O
this	O
model	O
,	O
when	O
the	O
initialization	B-KEY
procedure	I-KEY
is	O
defined	O
in	O
the	O
RPC	B-KEY
module	I-KEY
,	O
the	O
module	O
is	O
automatically	O
initialized	O
at	O
the	O
time	O
of	O
invocation	O
by	O
calling	O
the	O
initialization	B-KEY
procedure	I-KEY
.	O
This	O
can	O
eliminate	O
unnecessary	O
communication	O
and	O
initialization	O
at	O
each	O
call	O
in	O
CONFLEX-G	B-KEY
.	O
CONFLEXG	O
can	O
achieve	O
performance	O
comparable	O
to	O
CONFLEX	O
MPI	O
and	O
can	O
exploit	O
more	O
computing	O
resources	O
by	O
allowing	O
the	O
use	O
of	O
a	O
cluster	O
of	O
multiple	O
clusters	O
in	O
the	O
grid	O
.	O
The	O
experimental	O
result	O
shows	O
that	O
CONFLEX-G	B-KEY
achieved	O
a	O
speedup	O
of	O
56.5	O
times	O
in	O
the	O
case	O
of	O
the	O
1BL1	O
molecule	O
,	O
where	O
the	O
molecule	O
consists	O
of	O
a	O
large	O
number	O
of	O
atoms	O
,	O
and	O
each	O
trial	O
structure	O
optimization	O
requires	O
significant	O
time	O
.	O
The	O
load	O
imbalance	O
of	O
the	O
optimization	O
time	O
of	O
the	O
trial	O
structure	O
may	O
also	O
cause	O
performance	O
degradation	O
.	O

Computation	O
in	O
a	O
Distributed	B-KEY
Information	I-KEY
Market	O
∗	O
ABSTRACT	O
According	O
to	O
economic	B-KEY
theory	I-KEY
--	O
supported	O
by	O
empirical	B-KEY
and	I-KEY
laboratory	I-KEY
evidence	I-KEY
--	O
the	O
equilibrium	B-KEY
price	I-KEY
of	O
a	O
financial	B-KEY
security	I-KEY
reflects	O
all	O
of	O
the	O
information	O
regarding	O
the	O
security	O
's	O
value	O
.	O
We	O
investigate	O
the	O
computational	B-KEY
process	I-KEY
on	O
the	O
path	B-KEY
toward	I-KEY
equilibrium	I-KEY
,	O
where	O
information	O
distributed	O
among	O
traders	B-KEY
is	O
revealed	O
step-by-step	O
over	O
time	O
and	O
incorporated	O
into	O
the	O
market	B-KEY
price	I-KEY
.	O
We	O
develop	O
a	O
simplified	B-KEY
model	I-KEY
of	O
an	O
information	B-KEY
market	I-KEY
,	O
along	O
with	O
trading	B-KEY
strategies	I-KEY
,	O
in	O
order	O
to	O
formalize	O
the	O
computational	B-KEY
properties	I-KEY
of	I-KEY
the	I-KEY
process	I-KEY
.	O
We	O
show	O
that	O
securities	B-KEY
whose	O
payoffs	B-KEY
can	O
not	O
be	O
expressed	O
as	O
weighted	O
threshold	B-KEY
functions	I-KEY
of	O
distributed	O
input	O
bits	O
are	O
not	O
guaranteed	O
to	O
converge	O
to	O
the	O
proper	O
equilibrium	O
predicted	O
by	O
economic	B-KEY
theory	I-KEY
.	O
On	O
the	O
other	O
hand	O
,	O
securities	B-KEY
whose	O
payoffs	B-KEY
are	O
threshold	B-KEY
functions	I-KEY
are	O
guaranteed	O
to	O
converge	O
,	O
for	O
all	O
prior	O
probability	B-KEY
distributions	I-KEY
.	O
Moreover	O
,	O
these	O
threshold	O
securities	B-KEY
converge	O
in	O
at	O
most	O
n	O
rounds	B-KEY
,	O
where	O
n	O
is	O
the	O
number	B-KEY
of	I-KEY
bits	I-KEY
of	O
distributed	B-KEY
information	I-KEY
.	O
We	O
also	O
prove	O
a	O
lower	B-KEY
bound	I-KEY
,	O
showing	O
a	O
type	O
of	O
threshold	O
security	B-KEY
that	O
requires	O
at	O
least	O
n/2	O
rounds	B-KEY
to	O
converge	O
in	O
the	O
worst	B-KEY
case	I-KEY
.	O
∗	O
This	O
work	O
was	O
supported	O
by	O
the	O
DoD	O
University	O
Research	O
Initiative	O
-LRB-	O
URI	O
-RRB-	O
administered	O
by	O
the	O
Office	O
of	O
Naval	O
Research	O
under	O
Grant	O
N00014-01-1-0795	O
.	O
†	O
Supported	O
in	O
part	O
by	O
ONR	O
grant	O
N00014-01-0795	O
and	O
NSF	O
grants	O
CCR-0105337	O
,	O
CCR-TC-0208972	O
,	O
ANI-0207399	O
,	O
and	O
ITR-0219018	O
.	O
‡	O
This	O
work	O
conducted	O
while	O
at	O
NEC	O
Laboratories	O
America	O
,	O
Princeton	O
,	O
NJ	O
.	O

Nash	O
Equilibria	O
in	O
Graphical	B-KEY
Games	I-KEY
on	O
Trees	O
Revisited	O
*	O
Graphical	B-KEY
games	I-KEY
have	O
been	O
proposed	O
as	O
a	O
game-theoretic	O
model	O
of	O
large-scale	O
distributed	O
networks	O
of	O
non-cooperative	O
agents	O
.	O
When	O
the	O
number	O
of	O
players	O
is	O
large	O
,	O
and	O
the	O
underlying	O
graph	O
has	O
low	O
degree	B-KEY
,	O
they	O
provide	O
a	O
concise	O
way	O
to	O
represent	O
the	O
players	O
'	O
payoffs	O
.	O
It	O
has	O
recently	O
been	O
shown	O
that	O
the	O
problem	O
of	O
finding	O
Nash	O
equilibria	O
in	O
a	O
general	O
degree-3	O
graphical	B-KEY
game	I-KEY
with	O
two	O
actions	O
per	O
player	O
is	O
complete	O
for	O
the	O
complexity	O
class	O
PPAD	O
,	O
indicating	O
that	O
it	O
is	O
unlikely	O
that	O
there	O
is	O
any	O
polynomial-time	O
algorithm	O
for	O
this	O
problem	O
.	O
In	O
this	O
paper	O
,	O
we	O
study	O
the	O
complexity	O
of	O
graphical	B-KEY
games	I-KEY
with	O
two	O
actions	O
per	O
player	O
on	O
bounded-degree	O
trees	O
.	O
This	O
setting	O
was	O
first	O
considered	O
by	O
Kearns	O
,	O
Littman	O
and	O
Singh	O
,	O
who	O
proposed	O
a	O
dynamic	O
programming-based	O
algorithm	O
that	O
computes	O
all	O
Nash	O
equilibria	O
of	O
such	O
games	O
.	O
The	O
running	O
time	O
of	O
their	O
algorithm	O
is	O
exponential	O
,	O
though	O
approximate	O
equilibria	O
can	O
be	O
computed	O
efficiently	O
.	O
Later	O
,	O
Littman	O
,	O
Kearns	O
and	O
Singh	O
proposed	O
a	O
modification	O
to	O
this	O
algorithm	O
that	O
can	O
find	O
a	O
single	O
Nash	B-KEY
equilibrium	I-KEY
in	O
polynomial	O
time	O
.	O
We	O
show	O
that	O
this	O
modified	O
algorithm	O
is	O
incorrect	O
--	O
the	O
output	O
is	O
not	O
always	O
a	O
Nash	B-KEY
equilibrium	I-KEY
.	O
We	O
then	O
propose	O
a	O
new	O
algorithm	O
that	O
is	O
based	O
on	O
the	O
ideas	O
of	O
Kearns	O
et	O
al.	O
and	O
computes	O
all	O
Nash	O
equilibria	O
in	O
quadratic	O
time	O
if	O
the	O
input	O
graph	O
is	O
a	O
path	O
,	O
and	O
in	O
polynomial	O
time	O
if	O
it	O
is	O
an	O
arbitrary	O
graph	O
of	O
maximum	O
degree	B-KEY
2	O
.	O
Moreover	O
,	O
our	O
algorithm	O
can	O
be	O
used	O
to	O
compute	O
Nash	O
equilibria	O
of	O
graphical	B-KEY
games	I-KEY
on	O
arbitrary	O
trees	O
,	O
but	O
the	O
running	O
time	O
can	O
be	O
exponential	O
,	O
even	O
when	O
the	O
tree	O
has	O
bounded	O
degree	B-KEY
.	O
We	O
show	O
that	O
this	O
is	O
inevitable	O
--	O
any	O
algorithm	O
of	O
this	O
type	O
will	O
take	O
exponential	O
time	O
,	O
even	O
on	O
bounded-degree	O
trees	O
with	O
pathwidth	O
2	O
.	O
It	O
is	O
an	O
open	O
question	O
whether	O
our	O
algorithm	O
runs	O
in	O
polynomial	O
time	O
on	O
graphs	O
with	O
pathwidth	O
1	O
,	O
but	O
we	O
show	O
that	O
finding	O
a	O
Nash	B-KEY
equilibrium	I-KEY
for	O
a	O
2-action	O
graphical	B-KEY
game	I-KEY
in	O
which	O
the	O
underlying	O
graph	O
has	O
maximum	O
degree	B-KEY
3	O
and	O
constant	O
pathwidth	O
is	O
PPAD-complete	B-KEY
-LRB-	O
so	O
is	O
unlikely	O
to	O
be	O
tractable	O
-RRB-	O
.	O
*	O
This	O
research	O
is	O
supported	O
by	O
the	O
EPSRC	O
research	O
grants	O
``	O
Algorithmics	O
of	O
Network-sharing	O
Games	O
''	O
and	O
``	O
Discontinuous	O
Behaviour	O
in	O
the	O
Complexity	O
of	O
randomized	O
Algorithms	O
''	O
.	O

Implementation	B-KEY
with	O
a	O
Bounded	B-KEY
Action	I-KEY
Space	I-KEY
ABSTRACT	O
While	O
traditional	O
mechanism	O
design	O
typically	O
assumes	O
isomorphism	O
between	O
the	O
agents	O
'	O
type	O
-	O
and	O
action	O
spaces	O
,	O
in	O
many	O
situations	O
the	O
agents	O
face	O
strict	O
restrictions	O
on	O
their	O
action	O
space	O
due	O
to	O
,	O
e.g.	O
,	O
technical	O
,	O
behavioral	O
or	O
regulatory	O
reasons	O
.	O
We	O
devise	O
a	O
general	O
framework	O
for	O
the	O
study	O
of	O
mechanism	O
design	O
in	O
single-parameter	O
environments	O
with	O
restricted	O
action	O
spaces	O
.	O
Our	O
contribution	O
is	O
threefold	O
.	O
First	O
,	O
we	O
characterize	O
sufficient	O
conditions	O
under	O
which	O
the	O
information-theoretically	O
optimal	O
social-choice	O
rule	O
can	O
be	O
implemented	B-KEY
in	O
dominant	B-KEY
strategies	I-KEY
,	O
and	O
prove	O
that	O
any	O
multilinear	O
social-choice	O
rule	O
is	O
dominant-strategy	O
implementable	B-KEY
with	O
no	O
additional	O
cost	O
.	O
Second	O
,	O
we	O
identify	O
necessary	O
conditions	O
for	O
the	O
optimality	O
of	O
action-bounded	B-KEY
mechanisms	I-KEY
,	O
and	O
fully	O
characterize	O
the	O
optimal	B-KEY
mechanisms	I-KEY
and	O
strategies	O
in	O
games	O
with	O
two	O
players	O
and	O
two	O
alternatives	O
.	O
Finally	O
,	O
we	O
prove	O
that	O
for	O
any	O
multilinear	O
social-choice	O
rule	O
,	O
the	O
optimal	B-KEY
mechanism	I-KEY
with	O
k	O
actions	O
incurs	O
an	O
expected	O
loss	O
of	O
O	O
-LRB-	O
k21	O
-RRB-	O
compared	O
to	O
the	O
optimal	B-KEY
mechanisms	I-KEY
with	O
unrestricted	O
action	O
spaces	O
.	O
Our	O
results	O
apply	O
to	O
various	O
economic	O
and	O
computational	O
settings	O
,	O
and	O
we	O
demonstrate	O
their	O
applicability	O
to	O
signaling	O
games	O
,	O
public-good	O
models	O
and	O
routing	O
in	O
networks	O
.	O

Sequential	O
Decision	B-KEY
Making	O
in	O
Parallel	O
Two-Sided	O
Economic	O
Search	O
ABSTRACT	O
This	O
paper	O
presents	O
a	O
two-sided	O
economic	O
search	O
model	O
in	O
which	O
agents	O
are	O
searching	O
for	O
beneficial	O
pairwise	B-KEY
partnerships	I-KEY
.	O
In	O
each	O
search	O
stage	O
,	O
each	O
of	O
the	O
agents	O
is	O
randomly	O
matched	B-KEY
with	O
several	O
other	O
agents	O
in	O
parallel	O
,	O
and	O
makes	O
a	O
decision	B-KEY
whether	O
to	O
accept	O
a	O
potential	O
partnership	B-KEY
with	O
one	O
of	O
them	O
.	O
The	O
distinguishing	O
feature	O
of	O
the	O
proposed	O
model	O
is	O
that	O
the	O
agents	O
are	O
not	O
restricted	O
to	O
maintaining	O
a	O
synchronized	O
-LRB-	O
instantaneous	O
-RRB-	O
decision	B-KEY
protocol	O
and	O
can	O
sequentially	O
accept	O
and	O
reject	O
partnerships	B-KEY
within	O
the	O
same	O
search	O
stage	O
.	O
We	O
analyze	O
the	O
dynamics	O
which	O
drive	O
the	O
agents	O
'	O
strategies	O
towards	O
a	O
stable	O
equilibrium	O
in	O
the	O
new	O
model	O
and	O
show	O
that	O
the	O
proposed	O
search	O
strategy	O
weakly	O
dominates	O
the	O
one	O
currently	O
in	O
use	O
for	O
the	O
two-sided	O
parallel	O
economic	O
search	O
model	O
.	O
By	O
identifying	O
several	O
unique	O
characteristics	O
of	O
the	O
equilibrium	O
we	O
manage	O
to	O
efficiently	O
bound	O
the	O
strategy	O
space	O
that	O
needs	O
to	O
be	O
explored	O
by	O
the	O
agents	O
and	O
propose	O
an	O
efficient	O
means	O
for	O
extracting	O
the	O
distributed	O
equilibrium	B-KEY
strategies	I-KEY
in	O
common	O
environments	O
.	O

SMILE	O
:	O
Sound	O
Multi-agent	O
Incremental	B-KEY
LEarning	I-KEY
;--RRB-	O
*	O
ABSTRACT	O
This	O
article	O
deals	O
with	O
the	O
problem	O
of	O
collaborative	O
learning	O
in	O
a	O
multi-agent	O
system	O
.	O
Here	O
each	O
agent	B-KEY
can	O
update	O
incrementally	O
its	O
beliefs	O
B	O
-LRB-	O
the	O
concept	O
representation	O
-RRB-	O
so	O
that	O
it	O
is	O
in	O
a	O
way	O
kept	O
consistent	O
with	O
the	O
whole	O
set	O
of	O
information	O
K	O
-LRB-	O
the	O
examples	O
-RRB-	O
that	O
he	O
has	O
received	O
from	O
the	O
environment	O
or	O
other	O
agents	B-KEY
.	O
We	O
extend	O
this	O
notion	O
of	O
consistency	O
-LRB-	O
or	O
soundness	O
-RRB-	O
to	O
the	O
whole	O
MAS	O
and	O
discuss	O
how	O
to	O
obtain	O
that	O
,	O
at	O
any	O
moment	O
,	O
a	O
same	O
consistent	O
concept	O
representation	O
is	O
present	O
in	O
each	O
agent	B-KEY
.	O
The	O
corresponding	O
protocol	O
is	O
applied	O
to	O
supervised	O
concept	O
learning	O
.	O
The	O
resulting	O
method	O
SMILE	O
-LRB-	O
standing	O
for	O
Sound	O
Multiagent	O
Incremental	B-KEY
LEarning	I-KEY
-RRB-	O
is	O
described	O
and	O
experimented	O
here	O
.	O
Surprisingly	O
some	O
difficult	O
boolean	O
formulas	O
are	O
better	O
learned	O
,	O
given	O
the	O
same	O
learning	O
set	O
,	O
by	O
a	O
Multi	O
agent	B-KEY
system	O
than	O
by	O
a	O
single	O
agent	B-KEY
.	O

PackageBLAST	B-KEY
:	O
An	O
Adaptive	B-KEY
Multi-Policy	I-KEY
Grid	I-KEY
Service	I-KEY
for	O
Biological	B-KEY
Sequence	I-KEY
Comparison	I-KEY
*	O
ABSTRACT	O
In	O
this	O
paper	O
,	O
we	O
propose	O
an	O
adaptive	O
task	B-KEY
allocation	I-KEY
framework	O
to	O
perform	O
BLAST	B-KEY
searches	I-KEY
in	O
a	O
grid	B-KEY
environment	I-KEY
against	O
sequence	O
database	O
segments	O
.	O
The	O
framework	O
,	O
called	O
PackageBLAST	B-KEY
,	O
provides	O
an	O
infrastructure	O
to	O
choose	O
or	O
incorporate	O
task	B-KEY
allocation	I-KEY
strategies	O
.	O
Furthermore	O
,	O
we	O
propose	O
a	O
mechanism	O
to	O
compute	O
grid	O
nodes	O
execution	O
weight	O
,	O
adapting	O
the	O
chosen	O
allocation	O
policy	O
to	O
the	O
current	O
computational	O
power	O
of	O
the	O
nodes	O
.	O
Our	O
results	O
present	O
very	O
good	O
speedups	O
and	O
also	O
show	O
that	O
no	O
single	O
allocation	O
strategy	O
is	O
able	O
to	O
achieve	O
the	O
lowest	O
execution	O
times	O
for	O
all	O
scenarios	O
.	O

Analyzing	O
Feature	O
Trajectories	O
for	O
Event	B-KEY
Detection	I-KEY
ABSTRACT	O
We	O
consider	O
the	O
problem	O
of	O
analyzing	O
word	B-KEY
trajectories	I-KEY
in	O
both	O
time	O
and	O
frequency	O
domains	O
,	O
with	O
the	O
specific	O
goal	O
of	O
identifying	O
important	O
and	O
less-reported	O
,	O
periodic	O
and	O
aperiodic	O
words	O
.	O
A	O
set	O
of	O
words	O
with	O
identical	O
trends	O
can	O
be	O
grouped	O
together	O
to	O
reconstruct	O
an	O
event	O
in	O
a	O
completely	O
unsupervised	O
manner	O
.	O
The	O
document	O
frequency	O
of	O
each	O
word	O
across	O
time	O
is	O
treated	O
like	O
a	O
time	B-KEY
series	I-KEY
,	O
where	O
each	O
element	O
is	O
the	O
document	O
frequency	O
-	O
inverse	O
document	O
frequency	O
-LRB-	O
DFIDF	O
-RRB-	O
score	O
at	O
one	O
time	O
point	O
.	O
In	O
this	O
paper	O
,	O
we	O
1	O
-RRB-	O
first	O
applied	O
spectral	B-KEY
analysis	I-KEY
to	O
categorize	O
features	O
for	O
different	O
event	O
characteristics	O
:	O
important	O
and	O
less-reported	O
,	O
periodic	O
and	O
aperiodic	O
;	O
2	O
-RRB-	O
modeled	O
aperiodic	O
features	O
with	O
Gaussian	B-KEY
density	O
and	O
periodic	O
features	O
with	O
Gaussian	B-KEY
mixture	O
densities	O
,	O
and	O
subsequently	O
detected	O
each	O
feature	O
's	O
burst	O
by	O
the	O
truncated	O
Gaussian	B-KEY
approach	O
;	O
3	O
-RRB-	O
proposed	O
an	O
unsupervised	O
greedy	O
event	B-KEY
detection	I-KEY
algorithm	O
to	O
detect	O
both	O
aperiodic	O
and	O
periodic	B-KEY
events	I-KEY
.	O
All	O
of	O
the	O
above	O
methods	O
can	O
be	O
applied	O
to	O
time	B-KEY
series	I-KEY
data	O
in	O
general	O
.	O
We	O
extensively	O
evaluated	O
our	O
methods	O
on	O
the	O
1-year	O
Reuters	O
News	O
Corpus	O
-LSB-	O
3	O
-RSB-	O
and	O
showed	O
that	O
they	O
were	O
able	O
to	O
uncover	O
meaningful	O
aperiodic	O
and	O
periodic	B-KEY
events	I-KEY
.	O

Studying	O
the	O
Use	O
of	O
Popular	B-KEY
Destinations	I-KEY
to	O
Enhance	B-KEY
Web	I-KEY
Search	I-KEY
Interaction	O
ABSTRACT	O
We	O
present	O
a	O
novel	O
Web	B-KEY
search	I-KEY
interaction	I-KEY
feature	O
which	O
,	O
for	O
a	O
given	O
query	O
,	O
provides	O
links	O
to	O
websites	O
frequently	O
visited	O
by	O
other	O
users	O
with	O
similar	O
information	O
needs	O
.	O
These	O
popular	B-KEY
destinations	I-KEY
complement	O
traditional	O
search	O
results	O
,	O
allowing	O
direct	O
navigation	O
to	O
authoritative	O
resources	O
for	O
the	O
query	O
topic	O
.	O
Destinations	O
are	O
identified	O
using	O
the	O
history	O
of	O
search	O
and	O
browsing	O
behavior	O
of	O
many	O
users	O
over	O
an	O
extended	O
time	O
period	O
,	O
whose	O
collective	O
behavior	O
provides	O
a	O
basis	O
for	O
computing	O
source	O
authority	O
.	O
We	O
describe	O
a	O
user	B-KEY
study	I-KEY
which	O
compared	O
the	O
suggestion	O
of	O
destinations	O
with	O
the	O
previously	O
proposed	O
suggestion	O
of	O
related	B-KEY
queries	I-KEY
,	O
as	O
well	O
as	O
with	O
traditional	O
,	O
unaided	O
Web	O
search	O
.	O
Results	O
show	O
that	O
search	O
enhanced	O
by	O
destination	O
suggestions	O
outperforms	O
other	O
systems	O
for	O
exploratory	O
tasks	O
,	O
with	O
best	O
performance	O
obtained	O
from	O
mining	O
past	O
user	O
behavior	O
at	O
query-level	O
granularity	O
.	O

Scalable	O
Grid	B-KEY
Service	I-KEY
Discovery	I-KEY
Based	O
on	O
UDDI	O
*	O
ABSTRACT	O
Efficient	O
discovery	B-KEY
of	O
grid	O
services	O
is	O
essential	O
for	O
the	O
success	O
of	O
grid	B-KEY
computing	I-KEY
.	O
The	O
standardization	O
of	O
grids	O
based	O
on	O
web	B-KEY
services	I-KEY
has	O
resulted	O
in	O
the	O
need	O
for	O
scalable	O
web	B-KEY
service	I-KEY
discovery	B-KEY
mechanisms	O
to	O
be	O
deployed	O
in	O
grids	O
Even	O
though	O
UDDI	B-KEY
has	O
been	O
the	O
de	O
facto	O
industry	O
standard	O
for	O
web-services	O
discovery	B-KEY
,	O
imposed	O
requirements	O
of	O
tight-replication	O
among	O
registries	O
and	O
lack	O
of	O
autonomous	B-KEY
control	I-KEY
has	O
severely	O
hindered	O
its	O
widespread	O
deployment	O
and	O
usage	O
.	O
With	O
the	O
advent	O
of	O
grid	B-KEY
computing	I-KEY
the	O
scalability	B-KEY
issue	I-KEY
of	O
UDDI	B-KEY
will	O
become	O
a	O
roadblock	O
that	O
will	O
prevent	O
its	O
deployment	O
in	O
grids	O
.	O
In	O
this	O
paper	O
we	O
present	O
our	O
distributed	O
web-service	O
discovery	B-KEY
architecture	O
,	O
called	O
DUDE	O
-LRB-	O
Distributed	O
UDDI	O
Deployment	O
Engine	O
-RRB-	O
.	O
DUDE	O
leverages	O
DHT	B-KEY
-LRB-	O
Distributed	O
Hash	O
Tables	O
-RRB-	O
as	O
a	O
rendezvous	O
mechanism	O
between	O
multiple	O
UDDI	B-KEY
registries	O
.	O
DUDE	O
enables	O
consumers	O
to	O
query	B-KEY
multiple	O
registries	O
,	O
still	O
at	O
the	O
same	O
time	O
allowing	O
organizations	O
to	O
have	O
autonomous	B-KEY
control	I-KEY
over	O
their	O
registries	O
.	O
.	O
Based	O
on	O
preliminary	O
prototype	O
on	O
PlanetLab	O
,	O
we	O
believe	O
that	O
DUDE	O
architecture	O
can	O
support	O
effective	O
distribution	O
of	O
UDDI	B-KEY
registries	O
thereby	O
making	O
UDDI	O
more	O
robust	O
and	O
also	O
addressing	O
its	O
scaling	O
issues	O
.	O
Furthermore	O
,	O
The	O
DUDE	O
architecture	O
for	O
scalable	O
distribution	O
can	O
be	O
applied	O
beyond	O
UDDI	B-KEY
to	O
any	O
Grid	B-KEY
Service	I-KEY
Discovery	I-KEY
mechanism	O
.	O

An	O
Adversarial	B-KEY
Environment	I-KEY
Model	O
for	O
Bounded	O
Rational	O
Agents	B-KEY
in	O
Zero-Sum	O
Interactions	B-KEY
ABSTRACT	O
Multiagent	B-KEY
environments	I-KEY
are	O
often	O
not	O
cooperative	O
nor	O
collaborative	O
;	O
in	O
many	O
cases	O
,	O
agents	B-KEY
have	O
conflicting	O
interests	O
,	O
leading	O
to	O
adversarial	B-KEY
interactions	I-KEY
.	O
This	O
paper	O
presents	O
a	O
formal	O
Adversarial	B-KEY
Environment	I-KEY
model	O
for	O
bounded	O
rational	O
agents	B-KEY
operating	O
in	O
a	O
zero-sum	O
environment	O
.	O
In	O
such	O
environments	O
,	O
attempts	O
to	O
use	O
classical	O
utility-based	O
search	O
methods	O
can	O
raise	O
a	O
variety	O
of	O
difficulties	O
-LRB-	O
e.g.	O
,	O
implicitly	O
modeling	O
the	O
opponent	O
as	O
an	O
omniscient	O
utility	O
maximizer	O
,	O
rather	O
than	O
leveraging	O
a	O
more	O
nuanced	O
,	O
explicit	O
opponent	O
model	O
-RRB-	O
.	O
We	O
define	O
an	O
Adversarial	B-KEY
Environment	I-KEY
by	O
describing	O
the	O
mental	O
states	O
of	O
an	O
agent	B-KEY
in	O
such	O
an	O
environment	O
.	O
We	O
then	O
present	O
behavioral	B-KEY
axioms	I-KEY
that	O
are	O
intended	O
to	O
serve	O
as	O
design	O
principles	O
for	O
building	O
such	O
adversarial	O
agents	B-KEY
.	O
We	O
explore	O
the	O
application	O
of	O
our	O
approach	O
by	O
analyzing	O
log	O
files	O
of	O
completed	O
Connect-Four	B-KEY
games	I-KEY
,	O
and	O
present	O
an	O
empirical	O
analysis	O
of	O
the	O
axioms	O
'	O
appropriateness	O
.	O

Worst-Case	O
Optimal	O
Redistribution	O
of	O
VCG	O
Payments	O
in	O
Heterogeneous-Item	O
Auctions	O
with	O
Unit	O
Demand	O
ABSTRACT	O
Many	O
important	O
problems	O
in	O
multiagent	O
systems	O
involve	O
the	O
allocation	O
of	O
multiple	O
resources	O
among	O
the	O
agents	O
.	O
For	O
resource	O
allocation	O
problems	O
,	O
the	O
well-known	O
VCG	O
mechanism	B-KEY
satisfies	O
a	O
list	O
of	O
desired	O
properties	O
,	O
including	O
efficiency	O
,	O
strategy-proofness	O
,	O
individual	O
rationality	O
,	O
and	O
the	O
non-deficit	O
property	O
.	O
However	O
,	O
VCG	O
is	O
generally	O
not	O
budget-balanced	O
.	O
Under	O
VCG	O
,	O
agents	O
pay	O
the	O
VCG	O
payments	O
,	O
which	O
reduces	O
social	O
welfare	O
.	O
To	O
offset	O
the	O
loss	O
of	O
social	O
welfare	O
due	O
to	O
the	O
VCG	O
payments	O
,	O
VCG	O
redistribution	O
mechanisms	B-KEY
were	O
introduced	O
.	O
These	O
mechanisms	B-KEY
aim	O
to	O
redistribute	O
as	O
much	O
VCG	O
payments	O
back	O
to	O
the	O
agents	O
as	O
possible	O
,	O
while	O
maintaining	O
the	O
aforementioned	O
desired	O
properties	O
of	O
the	O
VCG	O
mechanism	B-KEY
.	O
We	O
continue	O
the	O
search	O
for	O
worst-case	O
optimal	O
VCG	O
redistribution	O
mechanisms	B-KEY
--	O
mechanisms	B-KEY
that	O
maximize	O
the	O
fraction	O
of	O
total	O
VCG	O
payment	B-KEY
redistributed	I-KEY
in	O
the	O
worst	O
case	O
.	O
Previously	O
,	O
a	O
worst-case	O
optimal	O
VCG	O
redistribution	O
mechanism	B-KEY
-LRB-	O
denoted	O
by	O
WCO	O
-RRB-	O
was	O
characterized	O
for	O
multi-unit	O
auctions	O
with	O
nonincreasing	O
marginal	O
values	O
-LSB-	O
7	O
-RSB-	O
.	O
Later	O
,	O
WCO	O
was	O
generalized	O
to	O
settings	O
involving	O
heterogeneous	O
items	O
-LSB-	O
4	O
-RSB-	O
,	O
resulting	O
in	O
the	O
HETERO	O
mechanism	B-KEY
.	O
-LSB-	O
4	O
-RSB-	O
conjectured	O
that	O
HETERO	O
is	O
feasible	O
and	O
worst-case	O
optimal	O
for	O
heterogeneous-item	O
auctions	O
with	O
unit	O
demand	O
.	O
In	O
this	O
paper	O
,	O
we	O
propose	O
a	O
more	O
natural	O
way	O
to	O
generalize	O
the	O
WCO	O
mechanism	B-KEY
.	O
We	O
prove	O
that	O
our	O
generalized	O
mechanism	B-KEY
,	O
though	O
represented	O
differently	O
,	O
actually	O
coincides	O
with	O
HETERO	O
.	O
Based	O
on	O
this	O
new	O
representation	O
of	O
HETERO	O
,	O
we	O
prove	O
that	O
HETERO	O
is	O
indeed	O
feasible	O
and	O
worst-case	O
optimal	O
in	O
heterogeneous-item	O
auctions	O
with	O
unit	O
demand	O
.	O
Finally	O
,	O
we	O
conjecture	O
that	O
HETERO	O
remains	O
feasible	O
and	O
worst-case	O
optimal	O
in	O
the	O
even	O
more	O
general	O
setting	O
of	O
combinatorial	O
auctions	O
with	O
gross	O
substitutes	O
.	O

Rewards-Based	O
Negotiation	B-KEY
for	O
Providing	O
Context	O
Information	O
ABSTRACT	O
How	O
to	O
provide	O
appropriate	O
context	O
information	O
is	O
a	O
challenging	O
problem	O
in	O
context-aware	B-KEY
computing	O
.	O
Most	O
existing	O
approaches	O
use	O
a	O
centralized	O
selection	O
mechanism	O
to	O
decide	O
which	O
context	O
information	O
is	O
appropriate	O
.	O
In	O
this	O
paper	O
,	O
we	O
propose	O
a	O
novel	O
approach	O
based	O
on	O
negotiation	B-KEY
with	O
rewards	O
to	O
solving	O
such	O
problem	O
.	O
Distributed	O
context	B-KEY
providers	I-KEY
negotiate	B-KEY
with	O
each	O
other	O
to	O
decide	O
who	O
can	O
provide	O
context	O
and	O
how	O
they	O
allocate	O
proceeds	O
.	O
In	O
order	O
to	O
support	O
our	O
approach	O
,	O
we	O
have	O
designed	O
a	O
concrete	O
negotiation	B-KEY
model	O
with	O
rewards	O
.	O
We	O
also	O
evaluate	O
our	O
approach	O
and	O
show	O
that	O
it	O
indeed	O
can	O
choose	O
an	O
appropriate	O
context	B-KEY
provider	I-KEY
and	O
allocate	O
the	O
proceeds	O
fairly	O
.	O

Topic	B-KEY
Segmentation	I-KEY
with	O
Shared	B-KEY
Topic	I-KEY
Detection	O
and	O
Alignment	O
of	O
Multiple	O
Documents	O
ABSTRACT	O
Topic	B-KEY
detection	I-KEY
and	O
tracking	B-KEY
-LSB-	O
26	O
-RSB-	O
and	O
topic	B-KEY
segmentation	I-KEY
-LSB-	O
15	O
-RSB-	O
play	O
an	O
important	O
role	O
in	O
capturing	O
the	O
local	B-KEY
and	I-KEY
sequential	I-KEY
information	I-KEY
of	I-KEY
documents	I-KEY
.	O
Previous	O
work	O
in	O
this	O
area	O
usually	O
focuses	O
on	O
single	B-KEY
documents	I-KEY
,	O
although	O
similar	O
multiple	B-KEY
documents	I-KEY
are	O
available	O
in	O
many	O
domains	O
.	O
In	O
this	O
paper	O
,	O
we	O
introduce	O
a	O
novel	O
unsupervised	O
method	O
for	O
shared	B-KEY
topic	I-KEY
detection	O
and	O
topic	O
segmentation	O
of	O
multiple	O
similar	O
documents	O
based	O
on	O
mutual	O
information	O
-LRB-	O
MI	O
-RRB-	O
and	O
weighted	O
mutual	O
information	O
-LRB-	O
WMI	O
-RRB-	O
that	O
is	O
a	O
combination	O
of	O
MI	O
and	O
term	O
weights	O
.	O
The	O
basic	O
idea	O
is	O
that	O
the	O
optimal	O
segmentation	O
maximizes	O
MI	O
-LRB-	O
or	O
WMI	O
-RRB-	O
.	O
Our	O
approach	O
can	O
detect	O
shared	B-KEY
topics	I-KEY
among	O
documents	O
.	O
It	O
can	O
find	O
the	O
optimal	B-KEY
boundaries	I-KEY
in	O
a	O
document	O
,	O
and	O
align	O
segments	O
among	O
documents	O
at	O
the	O
same	O
time	O
.	O
It	O
also	O
can	O
handle	O
single-document	O
segmentation	O
as	O
a	O
special	O
case	O
of	O
the	O
multi-document	O
segmentation	O
and	O
alignment	O
.	O
Our	O
methods	O
can	O
identify	O
and	O
strengthen	O
cue	B-KEY
terms	I-KEY
that	O
can	O
be	O
used	O
for	O
segmentation	O
and	O
partially	O
remove	O
stop	B-KEY
words	I-KEY
by	O
using	O
term	B-KEY
weights	I-KEY
based	O
on	O
entropy	O
learned	O
from	O
multiple	B-KEY
documents	I-KEY
.	O
Our	O
experimental	O
results	O
show	O
that	O
our	O
algorithm	O
works	O
well	O
for	O
the	O
tasks	O
of	O
single-document	O
segmentation	O
,	O
shared	B-KEY
topic	I-KEY
detection	O
,	O
and	O
multi-document	O
segmentation	O
.	O
Utilizing	O
information	O
from	O
multiple	B-KEY
documents	I-KEY
can	O
tremendously	O
improve	O
the	O
performance	B-KEY
of	I-KEY
topic	I-KEY
segmentation	I-KEY
,	O
and	O
using	O
WMI	O
is	O
even	O
better	O
than	O
using	O
MI	O
for	O
the	O
multi-document	O
segmentation	O
.	O

EDAS	B-KEY
:	O
Providing	O
an	O
Environment	O
for	O
Decentralized	O
Adaptive	B-KEY
Services	O
ABSTRACT	O
As	O
the	O
idea	O
of	O
virtualisation	O
of	O
compute	O
power	O
,	O
storage	O
and	O
bandwidth	O
becomes	O
more	O
and	O
more	O
important	O
,	O
grid	B-KEY
computing	I-KEY
evolves	O
and	O
is	O
applied	O
to	O
a	O
rising	O
number	O
of	O
applications	O
.	O
The	O
environment	O
for	O
decentralized	O
adaptive	B-KEY
services	O
-LRB-	O
EDAS	O
-RRB-	O
provides	O
a	O
grid-like	O
infrastructure	O
for	O
user-accessed	O
,	O
longterm	O
services	O
-LRB-	O
e.g.	O
webserver	O
,	O
source-code	O
repository	O
etc.	O
-RRB-	O
.	O
It	O
aims	O
at	O
supporting	O
the	O
autonomous	O
execution	O
and	O
evolution	O
of	O
services	O
in	O
terms	O
of	O
scalability	O
and	O
resource-aware	O
distribution	O
.	O
EDAS	B-KEY
offers	O
flexible	O
service	O
models	O
based	O
on	O
distributed	O
mobile	O
objects	O
ranging	O
from	O
a	O
traditional	O
clientserver	O
scenario	O
to	O
a	O
fully	O
peer-to-peer	O
based	O
approach	O
.	O
Automatic	O
,	O
dynamic	O
resource	B-KEY
management	O
allows	O
optimized	O
use	O
of	O
available	O
resources	O
while	O
minimizing	O
the	O
administrative	O
complexity	O
.	O

Distributed	O
Task	B-KEY
Allocation	I-KEY
in	O
Social	O
Networks	O
ABSTRACT	O
This	O
paper	O
proposes	O
a	O
new	O
variant	O
of	O
the	O
task	B-KEY
allocation	I-KEY
problem	O
,	O
where	O
the	O
agents	O
are	O
connected	O
in	O
a	O
social	O
network	O
and	O
tasks	O
arrive	O
at	O
the	O
agents	O
distributed	O
over	O
the	O
network	O
.	O
We	O
show	O
that	O
the	O
complexity	O
of	O
this	O
problem	O
remains	O
NPhard	O
.	O
Moreover	O
,	O
it	O
is	O
not	O
approximable	O
within	O
some	O
factor	O
.	O
We	O
develop	O
an	O
algorithm	B-KEY
based	O
on	O
the	O
contract-net	O
protocol	O
.	O
Our	O
algorithm	B-KEY
is	O
completely	O
distributed	O
,	O
and	O
it	O
assumes	O
that	O
agents	B-KEY
have	O
only	O
local	O
knowledge	O
about	O
tasks	O
and	O
resources	B-KEY
.	O
We	O
conduct	O
a	O
set	O
of	O
experiments	O
to	O
evaluate	O
the	O
performance	O
and	O
scalability	O
of	O
the	O
proposed	O
algorithm	B-KEY
in	O
terms	O
of	O
solution	O
quality	O
and	O
computation	O
time	O
.	O
Three	O
different	O
types	O
of	O
networks	O
,	O
namely	O
small-world	O
,	O
random	O
and	O
scale-free	O
networks	O
,	O
are	O
used	O
to	O
represent	O
various	O
social	B-KEY
relationships	I-KEY
among	O
agents	B-KEY
in	O
realistic	O
applications	O
.	O
The	O
results	O
demonstrate	O
that	O
our	O
algorithm	B-KEY
works	O
well	O
and	O
that	O
it	O
scales	O
well	O
to	O
large-scale	O
applications	O
.	O

Towards	O
Task-based	O
Personal	B-KEY
Information	I-KEY
Management	I-KEY
Evaluations	O
ABSTRACT	O
Personal	B-KEY
Information	I-KEY
Management	I-KEY
-LRB-	O
PIM	O
-RRB-	O
is	O
a	O
rapidly	O
growing	O
area	O
of	O
research	O
concerned	O
with	O
how	O
people	O
store	O
,	O
manage	O
and	O
re-find	B-KEY
information	I-KEY
.	O
A	O
feature	O
of	O
PIM	O
research	O
is	O
that	O
many	O
systems	O
have	O
been	O
designed	O
to	O
assist	O
users	O
manage	O
and	O
re-find	B-KEY
information	I-KEY
,	O
but	O
very	O
few	O
have	O
been	O
evaluated	O
.	O
This	O
has	O
been	O
noted	O
by	O
several	O
scholars	O
and	O
explained	O
by	O
the	O
difficulties	O
involved	O
in	O
performing	O
PIM	O
evaluations	O
.	O
The	O
difficulties	O
include	O
that	O
people	O
re-find	B-KEY
information	I-KEY
from	O
within	O
unique	O
personal	O
collections	O
;	O
researchers	O
know	O
little	O
about	O
the	O
tasks	O
that	O
cause	O
people	O
to	O
re-find	B-KEY
information	I-KEY
;	O
and	O
numerous	O
privacy	B-KEY
issues	I-KEY
concerning	O
personal	O
information	O
.	O
In	O
this	O
paper	O
we	O
aim	O
to	O
facilitate	O
PIM	O
evaluations	O
by	O
addressing	O
each	O
of	O
these	O
difficulties	O
.	O
In	O
the	O
first	O
part	O
,	O
we	O
present	O
a	O
diary	O
study	O
of	O
information	O
re-finding	O
tasks	O
.	O
The	O
study	O
examines	O
the	O
kind	O
of	O
tasks	O
that	O
require	O
users	O
to	O
re-find	B-KEY
information	I-KEY
and	O
produces	O
a	O
taxonomy	B-KEY
of	O
re-finding	O
tasks	O
for	O
email	B-KEY
messages	I-KEY
and	O
web	O
pages	O
.	O
In	O
the	O
second	O
part	O
,	O
we	O
propose	O
a	O
task-based	O
evaluation	O
methodology	O
based	O
on	O
our	O
findings	O
and	O
examine	O
the	O
feasibility	O
of	O
the	O
approach	O
using	O
two	O
different	O
methods	O
of	O
task	O
creation	O
.	O

Distributed	O
Management	B-KEY
of	O
Flexible	B-KEY
Times	I-KEY
Schedules	B-KEY
ABSTRACT	O
We	O
consider	O
the	O
problem	O
of	O
managing	B-KEY
schedules	O
in	O
an	O
uncertain	O
,	O
distributed	O
environment	O
.	O
We	O
assume	O
a	O
team	O
of	O
collaborative	O
agents	O
,	O
each	O
responsible	O
for	O
executing	O
a	O
portion	O
of	O
a	O
globally	O
pre-established	O
schedule	B-KEY
,	O
but	O
none	O
possessing	O
a	O
global	O
view	O
of	O
either	O
the	O
problem	O
or	O
solution	O
.	O
The	O
goal	O
is	O
to	O
maximize	O
the	O
joint	O
quality	O
obtained	O
from	O
the	O
activities	O
executed	O
by	O
all	O
agents	O
,	O
given	O
that	O
,	O
during	O
execution	O
,	O
unexpected	O
events	O
will	O
force	O
changes	O
to	O
some	O
prescribed	O
activities	O
and	O
reduce	O
the	O
utility	O
of	O
executing	O
others	O
.	O
We	O
describe	O
an	O
agent	B-KEY
architecture	I-KEY
for	O
solving	O
this	O
problem	O
that	O
couples	O
two	O
basic	O
mechanisms	O
:	O
-LRB-	O
1	O
-RRB-	O
a	O
``	O
flexible	B-KEY
times	I-KEY
''	O
representation	O
of	O
the	O
agent	O
's	O
schedule	B-KEY
-LRB-	O
using	O
a	O
Simple	O
Temporal	O
Network	O
-RRB-	O
and	O
-LRB-	O
2	O
-RRB-	O
an	O
incremental	O
rescheduling	O
procedure	O
.	O
The	O
former	O
hedges	O
against	O
temporal	O
uncertainty	O
by	O
allowing	O
execution	O
to	O
proceed	O
from	O
a	O
set	O
of	O
feasible	O
solutions	O
,	O
and	O
the	O
latter	O
acts	O
to	O
revise	O
the	O
agent	O
's	O
schedule	B-KEY
when	O
execution	O
is	O
forced	O
outside	O
of	O
this	O
set	O
of	O
solutions	O
or	O
when	O
execution	O
events	O
reduce	O
the	O
expected	O
value	O
of	O
this	O
feasible	O
solution	O
set	O
.	O
Basic	O
coordination	O
with	O
other	O
agents	O
is	O
achieved	O
simply	O
by	O
communicating	O
schedule	B-KEY
changes	O
to	O
those	O
agents	O
with	O
inter-dependent	B-KEY
activities	I-KEY
.	O
Then	O
,	O
as	O
time	O
permits	O
,	O
the	O
core	O
local	O
problem	O
solving	O
infra-structure	O
is	O
used	O
to	O
drive	O
an	O
inter-agent	O
option	O
generation	O
and	O
query	O
process	O
,	O
aimed	O
at	O
identifying	O
opportunities	O
for	O
solution	O
improvement	O
through	O
joint	O
change	O
.	O
Using	O
a	O
simulator	O
to	O
model	O
the	O
environment	O
,	O
we	O
compare	O
the	O
performance	B-KEY
of	O
our	O
multi-agent	O
system	O
with	O
that	O
of	O
an	O
expected	O
optimal	O
-LRB-	O
but	O
non-scalable	O
-RRB-	O
centralized	O
MDP	O
solver	O
.	O

Laplacian	O
Optimal	O
Design	O
for	O
Imag	O
e	O
Retrieval	O
ABSTRACT	O
Relevance	B-KEY
feedback	I-KEY
is	O
a	O
powerful	O
technique	O
to	O
enhance	O
ContentBased	B-KEY
Image	I-KEY
Retrieval	I-KEY
-LRB-	O
CBIR	O
-RRB-	O
performance	O
.	O
It	O
solicits	O
the	O
user	O
's	O
relevance	O
judgments	O
on	O
the	O
retrieved	O
images	O
returned	O
by	O
the	O
CBIR	O
systems	O
.	O
The	O
user	O
's	O
labeling	B-KEY
is	O
then	O
used	O
to	O
learn	O
a	O
classifier	O
to	O
distinguish	O
between	O
relevant	O
and	O
irrelevant	O
images	O
.	O
However	O
,	O
the	O
top	B-KEY
returned	I-KEY
images	I-KEY
may	O
not	O
be	O
the	O
most	O
informative	O
ones	O
.	O
The	O
challenge	O
is	O
thus	O
to	O
determine	O
which	O
unlabeled	O
images	O
would	O
be	O
the	O
most	O
informative	O
-LRB-	O
i.e.	O
,	O
improve	O
the	O
classifier	O
the	O
most	O
-RRB-	O
if	O
they	O
were	O
labeled	B-KEY
and	O
used	O
as	O
training	O
samples	O
.	O
In	O
this	O
paper	O
,	O
we	O
propose	O
a	O
novel	O
active	B-KEY
learning	I-KEY
algorithm	O
,	O
called	O
Laplacian	O
Optimal	O
Design	O
-LRB-	O
LOD	O
-RRB-	O
,	O
for	O
relevance	B-KEY
feedback	I-KEY
image	B-KEY
retrieval	I-KEY
.	O
Our	O
algorithm	O
is	O
based	O
on	O
a	O
regression	O
model	O
which	O
minimizes	O
the	O
least	O
square	O
error	O
on	O
the	O
measured	O
-LRB-	O
or	O
,	O
labeled	B-KEY
-RRB-	O
images	O
and	O
simultaneously	O
preserves	O
the	O
local	O
geometrical	O
structure	O
of	O
the	O
image	O
space	O
.	O
Specifically	O
,	O
we	O
assume	O
that	O
if	O
two	O
images	O
are	O
sufficiently	O
close	O
to	O
each	O
other	O
,	O
then	O
their	O
measurements	O
-LRB-	O
or	O
,	O
labels	B-KEY
-RRB-	O
are	O
close	O
as	O
well	O
.	O
By	O
constructing	O
a	O
nearest	O
neighbor	O
graph	O
,	O
the	O
geometrical	O
structure	O
of	O
the	O
image	O
space	O
can	O
be	O
described	O
by	O
the	O
graph	O
Laplacian	O
.	O
We	O
discuss	O
how	O
results	O
from	O
the	O
field	O
of	O
optimal	B-KEY
experimental	I-KEY
design	I-KEY
may	O
be	O
used	O
to	O
guide	O
our	O
selection	O
of	O
a	O
subset	O
of	O
images	O
,	O
which	O
gives	O
us	O
the	O
most	O
amount	O
of	O
information	O
.	O
Experimental	O
results	O
on	O
Corel	O
database	O
suggest	O
that	O
the	O
proposed	O
approach	O
achieves	O
higher	O
precision	O
in	O
relevance	B-KEY
feedback	I-KEY
image	B-KEY
retrieval	I-KEY
.	O

Apocrita	B-KEY
:	O
A	O
Distributed	O
Peer-to-Peer	O
File	B-KEY
Sharing	I-KEY
System	O
for	O
Intranets	O
ABSTRACT	O
Many	O
organizations	O
are	O
required	O
to	O
author	B-KEY
documents	B-KEY
for	O
various	O
purposes	O
,	O
and	O
such	O
documents	B-KEY
may	O
need	O
to	O
be	O
accessible	O
by	O
all	O
member	O
of	O
the	O
organization	O
.	O
This	O
access	O
may	O
be	O
needed	O
for	O
editing	O
or	O
simply	O
viewing	O
a	O
document	B-KEY
.	O
In	O
some	O
cases	O
these	O
documents	B-KEY
are	O
shared	O
between	O
authors	B-KEY
,	O
via	O
email	O
,	O
to	O
be	O
edited	O
.	O
This	O
can	O
easily	O
cause	O
incorrect	O
version	O
to	O
be	O
sent	O
or	O
conflicts	O
created	O
between	O
multiple	O
users	O
trying	O
to	O
make	O
amendments	O
to	O
a	O
document	B-KEY
.	O
There	O
may	O
even	O
be	O
multiple	O
different	O
documents	B-KEY
in	O
the	O
process	O
of	O
being	O
edited	O
.	O
The	O
user	O
may	O
be	O
required	O
to	O
search	O
for	O
a	O
particular	O
document	B-KEY
,	O
which	O
some	O
search	O
tools	O
such	O
as	O
Google	O
Desktop	O
may	O
be	O
a	O
solution	O
for	O
local	O
documents	B-KEY
but	O
will	O
not	O
find	O
a	O
document	B-KEY
on	O
another	O
user	O
's	O
machine	O
.	O
Another	O
problem	O
arises	O
when	O
a	O
document	B-KEY
is	O
made	O
available	O
on	O
a	O
user	O
's	O
machine	O
and	O
that	O
user	O
is	O
offline	O
,	O
in	O
which	O
case	O
the	O
document	B-KEY
is	O
no	O
longer	O
accessible	O
.	O
In	O
this	O
paper	O
we	O
present	O
Apocrita	B-KEY
,	O
a	O
revolutionary	O
distributed	O
P2P	B-KEY
file	B-KEY
sharing	I-KEY
system	O
for	O
Intranets	O
.	O

Mediators	B-KEY
in	O
Position	B-KEY
Auctions	I-KEY
ABSTRACT	O
A	O
mediator	B-KEY
is	O
a	O
reliable	O
entity	O
,	O
which	O
can	O
play	O
on	O
behalf	O
of	O
agents	B-KEY
in	O
a	O
given	O
game	O
.	O
A	O
mediator	B-KEY
however	O
can	O
not	O
enforce	O
the	O
use	O
of	O
its	O
services	O
,	O
and	O
each	O
agent	B-KEY
is	O
free	O
to	O
participate	O
in	O
the	O
game	O
directly	O
.	O
In	O
this	O
paper	O
we	O
introduce	O
a	O
study	O
of	O
mediators	B-KEY
for	O
games	O
with	O
incomplete	O
information	O
,	O
and	O
apply	O
it	O
to	O
the	O
context	O
of	O
position	B-KEY
auctions	I-KEY
,	O
a	O
central	O
topic	O
in	O
electronic	O
commerce	O
.	O
VCG	O
position	B-KEY
auctions	I-KEY
,	O
which	O
are	O
currently	O
not	O
used	O
in	O
practice	O
,	O
possess	O
some	O
nice	O
theoretical	O
properties	O
,	O
such	O
as	O
the	O
optimization	O
of	O
social	O
surplus	O
and	O
having	O
dominant	O
strategies	O
.	O
These	O
properties	O
may	O
not	O
be	O
satisfied	O
by	O
current	O
position	B-KEY
auctions	I-KEY
and	O
their	O
variants	O
.	O
We	O
therefore	O
concentrate	O
on	O
the	O
search	O
for	O
mediators	B-KEY
that	O
will	O
allow	O
to	O
transform	O
current	O
position	B-KEY
auctions	I-KEY
into	O
VCG	O
position	O
auctions	O
.	O
We	O
require	O
that	O
accepting	O
the	O
mediator	B-KEY
services	O
,	O
and	O
reporting	O
honestly	O
to	O
the	O
mediator	B-KEY
,	O
will	O
form	O
an	O
ex	B-KEY
post	I-KEY
equilibrium	I-KEY
,	O
which	O
satisfies	O
the	O
following	O
rationality	O
condition	O
:	O
an	O
agent	O
's	O
payoff	O
can	O
not	O
be	O
negative	O
regardless	O
of	O
the	O
actions	O
taken	O
by	O
the	O
agents	O
who	O
did	O
not	O
choose	O
the	O
mediator	O
's	O
services	O
,	O
or	O
by	O
the	O
agents	O
who	O
report	O
false	O
types	O
to	O
the	O
mediator	O
.	O
We	O
prove	O
the	O
existence	O
of	O
such	O
desired	O
mediators	B-KEY
for	O
the	O
next-price	O
-LRB-	O
Google-like	O
-RRB-	O
position	B-KEY
auctions	I-KEY
,	O
as	O
well	O
as	O
for	O
a	O
richer	O
class	O
of	O
position	O
auctions	O
,	O
including	O
all	O
k-price	O
position	O
auctions	O
,	O
k	O
>	O
1	O
.	O
For	O
k	O
=	O
1	O
,	O
the	O
self-price	O
position	B-KEY
auction	I-KEY
,	O
we	O
show	O
that	O
the	O
existence	O
of	O
such	O
mediator	O
depends	O
on	O
the	O
tie	O
breaking	O
rule	O
used	O
in	O
the	O
auction	O
.	O

A	O
Strategic	O
Model	O
for	O
Information	B-KEY
Markets	I-KEY
ABSTRACT	O
Information	B-KEY
markets	I-KEY
,	O
which	O
are	O
designed	O
specifically	O
to	O
aggregate	O
traders	O
'	O
information	O
,	O
are	O
becoming	O
increasingly	O
popular	O
as	O
a	O
means	O
for	O
predicting	O
future	O
events	O
.	O
Recent	O
research	O
in	O
information	B-KEY
markets	I-KEY
has	O
resulted	O
in	O
two	O
new	O
designs	O
,	O
market	B-KEY
scoring	I-KEY
rules	I-KEY
and	O
dynamic	B-KEY
parimutuel	I-KEY
markets	I-KEY
.	O
We	O
develop	O
an	O
analytic	O
method	O
to	O
guide	O
the	O
design	O
and	O
strategic	B-KEY
analysis	I-KEY
of	O
information	B-KEY
markets	I-KEY
.	O
Our	O
central	O
contribution	O
is	O
a	O
new	O
abstract	O
betting	O
game	O
,	O
the	O
projection	B-KEY
game	I-KEY
,	O
that	O
serves	O
as	O
a	O
useful	O
model	O
for	O
information	B-KEY
markets	I-KEY
.	O
We	O
demonstrate	O
that	O
this	O
game	O
can	O
serve	O
as	O
a	O
strategic	O
model	O
of	O
dynamic	B-KEY
parimutuel	I-KEY
markets	I-KEY
,	O
and	O
also	O
captures	O
the	O
essence	O
of	O
the	O
strategies	O
in	O
market	B-KEY
scoring	I-KEY
rules	I-KEY
.	O
The	O
projection	B-KEY
game	I-KEY
is	O
tractable	O
to	O
analyze	O
,	O
and	O
has	O
an	O
attractive	O
geometric	O
visualization	O
that	O
makes	O
the	O
strategic	O
moves	O
and	O
interactions	O
more	O
transparent	O
.	O
We	O
use	O
it	O
to	O
prove	O
several	O
strategic	O
properties	O
about	O
the	O
dynamic	B-KEY
parimutuel	I-KEY
market	I-KEY
.	O
We	O
also	O
prove	O
that	O
a	O
special	O
form	O
of	O
the	O
projection	B-KEY
game	I-KEY
is	O
strategically	O
equivalent	O
to	O
the	O
spherical	B-KEY
scoring	I-KEY
rule	I-KEY
,	O
and	O
it	O
is	O
strategically	O
similar	O
to	O
other	O
scoring	O
rules	O
.	O
Finally	O
,	O
we	O
illustrate	O
two	O
applications	O
of	O
the	O
model	O
to	O
analysis	O
of	O
complex	O
strategic	O
scenarios	O
:	O
we	O
analyze	O
the	O
precision	O
of	O
a	O
market	O
in	O
which	O
traders	O
have	O
inertia	O
,	O
and	O
a	O
market	O
in	O
which	O
a	O
trader	O
can	O
profit	O
by	O
manipulating	O
another	O
trader	O
's	O
beliefs	O
.	O

Aborting	O
Tasks	B-KEY
in	O
BDI	O
Agents	B-KEY
ABSTRACT	O
Intelligent	B-KEY
agents	I-KEY
that	O
are	O
intended	O
to	O
work	O
in	O
dynamic	O
environments	O
must	O
be	O
able	O
to	O
gracefully	O
handle	O
unsuccessful	O
tasks	O
and	O
plans	O
.	O
In	O
addition	O
,	O
such	O
agents	B-KEY
should	O
be	O
able	O
to	O
make	O
rational	O
decisions	O
about	O
an	O
appropriate	O
course	O
of	O
action	O
,	O
which	O
may	O
include	O
aborting	O
a	O
task	B-KEY
or	O
plan	O
,	O
either	O
as	O
a	O
result	O
of	O
the	O
agent	B-KEY
's	O
own	O
deliberations	O
,	O
or	O
potentially	O
at	O
the	O
request	O
of	O
another	O
agent	B-KEY
.	O
In	O
this	O
paper	O
we	O
investigate	O
the	O
incorporation	O
of	O
aborts	O
into	O
a	O
BDI-style	O
architecture	O
.	O
We	O
discuss	O
some	O
conditions	O
under	O
which	O
aborting	O
a	O
task	B-KEY
or	O
plan	O
is	O
appropriate	O
,	O
and	O
how	O
to	O
determine	O
the	O
consequences	O
of	O
such	O
a	O
decision	O
.	O
We	O
augment	O
each	O
plan	O
with	O
an	O
optional	O
abort-method	B-KEY
,	O
analogous	O
to	O
the	O
failure	B-KEY
method	O
found	O
in	O
some	O
agent	B-KEY
programming	O
languages	O
.	O
We	O
provide	O
an	O
operational	B-KEY
semantics	I-KEY
for	O
the	O
execution	O
cycle	O
in	O
the	O
presence	O
of	O
aborts	O
in	O
the	O
abstract	O
agent	B-KEY
language	O
CAN	O
,	O
which	O
enables	O
us	O
to	O
specify	O
a	O
BDI-based	O
execution	O
model	O
without	O
limiting	O
our	O
attention	O
to	O
a	O
particular	O
agent	B-KEY
system	O
-LRB-	O
such	O
as	O
JACK	O
,	O
Jadex	O
,	O
Jason	O
,	O
or	O
SPARK	O
-RRB-	O
.	O
A	O
key	O
technical	O
challenge	O
we	O
address	O
is	O
the	O
presence	O
of	O
parallel	O
execution	O
threads	O
and	O
of	O
sub-tasks	O
,	O
which	O
require	O
the	O
agent	O
to	O
ensure	O
that	O
the	O
abort	O
methods	O
for	O
each	O
plan	O
are	O
carried	O
out	O
in	O
an	O
appropriate	O
sequence	O
.	O

Bullet	B-KEY
:	O
High	O
Bandwidth	B-KEY
Data	B-KEY
Dissemination	I-KEY
Using	O
an	O
Overlay	O
Mesh	O
ABSTRACT	O
In	O
recent	O
years	O
,	O
overlay	O
networks	O
have	O
become	O
an	O
effective	O
alternative	O
to	O
IP	B-KEY
multicast	I-KEY
for	O
efficient	O
point	O
to	O
multipoint	B-KEY
communication	I-KEY
across	O
the	O
Internet	O
.	O
Typically	O
,	O
nodes	O
self-organize	O
with	O
the	O
goal	O
of	O
forming	O
an	O
efficient	O
overlay	O
tree	O
,	O
one	O
that	O
meets	O
performance	O
targets	O
without	O
placing	O
undue	O
burden	O
on	O
the	O
underlying	O
network	O
.	O
In	O
this	O
paper	O
,	O
we	O
target	O
high-bandwidth	O
data	O
distribution	O
from	O
a	O
single	O
source	O
to	O
a	O
large	O
number	O
of	O
receivers	O
.	O
Applications	O
include	O
large-file	O
transfers	O
and	O
real-time	O
multimedia	O
streaming	O
.	O
For	O
these	O
applications	O
,	O
we	O
argue	O
that	O
an	O
overlay	O
mesh	O
,	O
rather	O
than	O
a	O
tree	O
,	O
can	O
deliver	O
fundamentally	O
higher	O
bandwidth	B-KEY
and	O
reliability	O
relative	O
to	O
typical	O
tree	O
structures	O
.	O
This	O
paper	O
presents	O
Bullet	B-KEY
,	O
a	O
scalable	O
and	O
distributed	O
algorithm	O
that	O
enables	O
nodes	O
spread	O
across	O
the	O
Internet	O
to	O
self-organize	O
into	O
a	O
high	O
bandwidth	B-KEY
overlay	O
mesh	O
.	O
We	O
construct	O
Bullet	B-KEY
around	O
the	O
insight	O
that	O
data	O
should	O
be	O
distributed	O
in	O
a	O
disjoint	O
manner	O
to	O
strategic	O
points	O
in	O
the	O
network	O
.	O
Individual	O
Bullet	B-KEY
receivers	O
are	O
then	O
responsible	O
for	O
locating	O
and	O
retrieving	O
the	O
data	O
from	O
multiple	O
points	O
in	O
parallel	O
.	O
Key	O
contributions	O
of	O
this	O
work	O
include	O
:	O
i	O
-RRB-	O
an	O
algorithm	O
that	O
sends	O
data	O
to	O
different	O
points	O
in	O
the	O
overlay	O
such	O
that	O
any	O
data	O
object	O
is	O
equally	O
likely	O
to	O
appear	O
at	O
any	O
node	O
,	O
ii	O
-RRB-	O
a	O
scalable	O
and	O
decentralized	O
algorithm	O
that	O
allows	O
nodes	O
to	O
locate	O
and	O
recover	O
missing	O
data	O
items	O
,	O
and	O
iii	O
-RRB-	O
a	O
complete	O
implementation	O
and	O
evaluation	O
of	O
Bullet	B-KEY
running	O
across	O
the	O
Internet	O
and	O
in	O
a	O
large-scale	O
emulation	O
environment	O
reveals	O
up	O
to	O
a	O
factor	O
two	O
bandwidth	B-KEY
improvements	O
under	O
a	O
variety	O
of	O
circumstances	O
.	O
In	O
addition	O
,	O
we	O
find	O
that	O
,	O
relative	O
to	O
tree-based	O
solutions	O
,	O
Bullet	B-KEY
reduces	O
the	O
need	O
to	O
perform	O
expensive	O
bandwidth	B-KEY
probing	O
.	O
In	O
a	O
tree	O
,	O
it	O
is	O
critical	O
that	O
a	O
node	O
's	O
parent	O
delivers	O
a	O
high	O
rate	O
of	O
application	O
data	O
to	O
each	O
child	O
.	O
In	O
Bullet	B-KEY
however	O
,	O
nodes	O
simultaneously	O
receive	O
data	O
from	O
multiple	O
sources	O
in	O
parallel	O
,	O
making	O
it	O
less	O
important	O
to	O
locate	O
any	O
single	O
source	O
capable	O
of	O
sustaining	O
a	O
high	O
transmission	O
rate	O
.	O

Implementation	O
of	O
a	O
Dynamic	O
Adjustment	O
Mechanism	O
with	O
Efficient	O
Replica	B-KEY
Selection	O
in	O
Data	O
Grid	O
Environments	O
ABSTRACT	O
The	O
co-allocation	B-KEY
architecture	O
was	O
developed	O
in	O
order	O
to	O
enable	O
parallel	O
downloading	O
of	O
datasets	O
from	O
multiple	O
servers	B-KEY
.	O
Several	O
co-allocation	B-KEY
strategies	O
have	O
been	O
coupled	O
and	O
used	O
to	O
exploit	O
rate	O
differences	O
among	O
various	O
client-server	O
links	O
and	O
to	O
address	O
dynamic	O
rate	O
fluctuations	O
by	O
dividing	O
files	O
into	O
multiple	O
blocks	O
of	O
equal	O
sizes	O
.	O
However	O
,	O
a	O
major	O
obstacle	O
,	O
the	O
idle	O
time	O
of	O
faster	O
servers	B-KEY
having	O
to	O
wait	O
for	O
the	O
slowest	O
server	B-KEY
to	O
deliver	O
the	O
final	O
block	O
,	O
makes	O
it	O
important	O
to	O
reduce	O
differences	O
in	O
finishing	O
time	O
among	O
replica	B-KEY
servers	B-KEY
.	O
In	O
this	O
paper	O
,	O
we	O
propose	O
a	O
dynamic	O
coallocation	O
scheme	O
,	O
namely	O
Recursive-Adjustment	O
Co-Allocation	B-KEY
scheme	O
,	O
to	O
improve	O
the	O
performance	B-KEY
of	O
data	B-KEY
transfer	I-KEY
in	O
Data	B-KEY
Grids	I-KEY
.	O
Our	O
approach	O
reduces	O
the	O
idle	O
time	O
spent	O
waiting	O
for	O
the	O
slowest	O
server	B-KEY
and	O
decreases	O
data	B-KEY
transfer	I-KEY
completion	O
time	O
.	O
We	O
also	O
provide	O
an	O
effective	O
scheme	O
for	O
reducing	O
the	O
cost	O
of	O
reassembling	O
data	O
blocks	O
.	O

Using	O
Query	B-KEY
Contexts	I-KEY
in	O
Information	O
Retrieval	O
ABSTRACT	O
User	O
query	O
is	O
an	O
element	O
that	O
specifies	O
an	O
information	B-KEY
need	I-KEY
,	O
but	O
it	O
is	O
not	O
the	O
only	O
one	O
.	O
Studies	O
in	O
literature	O
have	O
found	O
many	O
contextual	O
factors	O
that	O
strongly	O
influence	O
the	O
interpretation	O
of	O
a	O
query	O
.	O
Recent	O
studies	O
have	O
tried	O
to	O
consider	O
the	O
user	O
's	O
interests	O
by	O
creating	O
a	O
user	B-KEY
profile	I-KEY
.	O
However	O
,	O
a	O
single	O
profile	O
for	O
a	O
user	O
may	O
not	O
be	O
sufficient	O
for	O
a	O
variety	O
of	O
queries	O
of	O
the	O
user	O
.	O
In	O
this	O
study	O
,	O
we	O
propose	O
to	O
use	O
query-specific	O
contexts	O
instead	O
of	O
user-centric	O
ones	O
,	O
including	O
context	O
around	O
query	O
and	O
context	O
within	O
query	O
.	O
The	O
former	O
specifies	O
the	O
environment	O
of	O
a	O
query	O
such	O
as	O
the	O
domain	B-KEY
of	I-KEY
interest	I-KEY
,	O
while	O
the	O
latter	O
refers	O
to	O
context	O
words	O
within	O
the	O
query	O
,	O
which	O
is	O
particularly	O
useful	O
for	O
the	O
selection	O
of	O
relevant	O
term	B-KEY
relations	I-KEY
.	O
In	O
this	O
paper	O
,	O
both	O
types	O
of	O
context	O
are	O
integrated	O
in	O
an	O
IR	O
model	O
based	O
on	O
language	B-KEY
modeling	I-KEY
.	O
Our	O
experiments	O
on	O
several	O
TREC	O
collections	O
show	O
that	O
each	O
of	O
the	O
context	B-KEY
factors	I-KEY
brings	O
significant	O
improvements	O
in	O
retrieval	O
effectiveness	O
.	O

Self-Adaptive	B-KEY
Applications	O
on	O
the	O
Grid	O
Abstract	O
Grids	O
are	O
inherently	O
heterogeneous	O
and	O
dynamic	O
.	O
One	O
important	O
problem	O
in	O
grid	B-KEY
computing	I-KEY
is	O
resource	B-KEY
selection	I-KEY
,	O
that	O
is	O
,	O
finding	O
an	O
appropriate	O
resource	O
set	O
for	O
the	O
application	O
.	O
Another	O
problem	O
is	O
adaptation	O
to	O
the	O
changing	O
characteristics	O
of	O
the	O
grid	B-KEY
environment	I-KEY
.	O
Existing	O
solutions	O
to	O
these	O
two	O
problems	O
require	O
that	O
a	O
performance	O
model	O
for	O
an	O
application	O
is	O
known	O
.	O
However	O
,	O
constructing	O
such	O
models	O
is	O
a	O
complex	O
task	O
.	O
In	O
this	O
paper	O
,	O
we	O
investigate	O
an	O
approach	O
that	O
does	O
not	O
require	O
performance	O
models	O
.	O
We	O
start	O
an	O
application	O
on	O
any	O
set	O
of	O
resources	O
.	O
During	O
the	O
application	O
run	O
,	O
we	O
periodically	O
collect	O
the	O
statistics	O
about	O
the	O
application	O
run	O
and	O
deduce	O
application	O
requirements	O
from	O
these	O
statistics	O
.	O
Then	O
,	O
we	O
adjust	O
the	O
resource	O
set	O
to	O
better	O
fit	O
the	O
application	O
needs	O
.	O
This	O
approach	O
allows	O
us	O
to	O
avoid	O
performance	O
bottlenecks	O
,	O
such	O
as	O
overloaded	O
WAN	O
links	O
or	O
very	O
slow	O
processors	O
,	O
and	O
therefore	O
can	O
yield	O
significant	O
performance	O
improvements	O
.	O
We	O
evaluate	O
our	O
approach	O
in	O
a	O
number	O
of	O
scenarios	O
typical	O
for	O
the	O
Grid	O
.	O

Truthful	B-KEY
Mechanism	I-KEY
Design	I-KEY
for	O
Multi-Dimensional	O
Scheduling	O
via	O
Cycle	O
Monotonicity	O
ABSTRACT	O
We	O
consider	O
the	O
problem	O
of	O
makespan	B-KEY
minimization	I-KEY
on	O
m	O
unrelated	O
machines	O
in	O
the	O
context	O
of	O
algorithmic	B-KEY
mechanism	B-KEY
design	I-KEY
,	O
where	O
the	O
machines	O
are	O
the	O
strategic	O
players	O
.	O
This	O
is	O
a	O
multidimensional	O
scheduling	B-KEY
domain	O
,	O
and	O
the	O
only	O
known	O
positive	O
results	O
for	O
makespan	B-KEY
minimization	I-KEY
in	O
such	O
a	O
domain	O
are	O
O	O
-LRB-	O
m	O
-RRB-	O
-	O
approximation	O
truthful	O
mechanisms	O
-LSB-	O
22	O
,	O
20	O
-RSB-	O
.	O
We	O
study	O
a	O
well-motivated	O
special	O
case	O
of	O
this	O
problem	O
,	O
where	O
the	O
processing	O
time	O
of	O
a	O
job	O
on	O
each	O
machine	O
may	O
either	O
be	O
``	O
low	O
''	O
or	O
``	O
high	O
''	O
,	O
and	O
the	O
low	O
and	O
high	O
values	O
are	O
public	O
and	O
job-dependent	O
.	O
This	O
preserves	O
the	O
multidimensionality	O
of	O
the	O
domain	O
,	O
and	O
generalizes	O
the	O
restricted-machines	O
-LRB-	O
i.e.	O
,	O
-LCB-	O
pj	O
,	O
∞	O
-RCB-	O
-RRB-	O
setting	O
in	O
scheduling	B-KEY
.	O
We	O
give	O
a	O
general	O
technique	O
to	O
convert	O
any	O
c-approximation	O
algorithm	O
to	O
a	O
3capproximation	O
truthful-in-expectation	O
mechanism	O
.	O
This	O
is	O
one	O
of	O
the	O
few	O
known	O
results	O
that	O
shows	O
how	O
to	O
export	O
approximation	B-KEY
algorithms	I-KEY
for	O
a	O
multidimensional	O
problem	O
into	O
truthful	O
mechanisms	O
in	O
a	O
black-box	O
fashion	O
.	O
When	O
the	O
low	O
and	O
high	O
values	O
are	O
the	O
same	O
for	O
all	O
jobs	O
,	O
we	O
devise	O
a	O
deterministic	O
2-approximation	O
truthful	O
mechanism	O
.	O
These	O
are	O
the	O
first	O
truthful	O
mechanisms	O
with	O
non-trivial	O
performance	O
guarantees	O
for	O
a	O
multidimensional	O
scheduling	B-KEY
domain	O
.	O
Our	O
constructions	O
are	O
novel	O
in	O
two	O
respects	O
.	O
First	O
,	O
we	O
do	O
not	O
utilize	O
or	O
rely	O
on	O
explicit	O
price	O
definitions	O
to	O
prove	O
truthfulness	O
;	O
instead	O
we	O
design	O
algorithms	B-KEY
that	O
satisfy	O
cycle	B-KEY
monotonicity	I-KEY
.	O
Cycle	B-KEY
monotonicity	I-KEY
-LSB-	O
23	O
-RSB-	O
is	O
a	O
necessary	O
and	O
sufficient	O
condition	O
for	O
truthfulness	O
,	O
is	O
a	O
generalization	O
of	O
value	O
monotonicity	O
for	O
multidimensional	O
domains	O
.	O
However	O
,	O
whereas	O
value	O
monotonicity	O
has	O
been	O
used	O
extensively	O
and	O
successfully	O
to	O
design	O
truthful	O
mechanisms	O
in	O
singledimensional	O
domains	O
,	O
ours	O
is	O
the	O
first	O
work	O
that	O
leverages	O
cycle	B-KEY
monotonicity	I-KEY
in	O
the	O
multidimensional	O
setting	O
.	O
Second	O
,	O
our	O
randomized	B-KEY
mechanisms	I-KEY
are	O
obtained	O
by	O
first	O
constructing	O
a	O
fractional	O
truthful	O
mechanism	O
for	O
a	O
fractional	O
relaxation	O
of	O
the	O
problem	O
,	O
and	O
then	O
converting	O
it	O
into	O
a	O
truthfulin-expectation	O
mechanism	O
.	O
This	O
builds	O
upon	O
a	O
technique	O
of	O
-LSB-	O
16	O
-RSB-	O
,	O
and	O
shows	O
the	O
usefulness	O
of	O
fractional	O
mechanisms	O
in	O
truthful	B-KEY
mechanism	I-KEY
design	I-KEY
.	O

Frugality	B-KEY
Ratios	O
And	O
Improved	O
Truthful	O
Mechanisms	O
for	O
Vertex	O
Cover	O
*	O
In	O
set-system	O
auctions	B-KEY
,	O
there	O
are	O
several	O
overlapping	O
teams	O
of	O
agents	O
,	O
and	O
a	O
task	O
that	O
can	O
be	O
completed	O
by	O
any	O
of	O
these	O
teams	O
.	O
The	O
auctioneer	B-KEY
's	O
goal	O
is	O
to	O
hire	O
a	O
team	O
and	O
pay	O
as	O
little	O
as	O
possible	O
.	O
Examples	O
of	O
this	O
setting	O
include	O
shortest-path	O
auctions	B-KEY
and	O
vertex-cover	O
auctions	B-KEY
.	O
Recently	O
,	O
Karlin	O
,	O
Kempe	O
and	O
Tamir	O
introduced	O
a	O
new	O
definition	O
offrugality	O
ratio	O
for	O
this	O
problem	O
.	O
Informally	O
,	O
the	O
``	O
frugality	B-KEY
ratio	O
''	O
is	O
the	O
ratio	O
of	O
the	O
total	O
payment	O
of	O
a	O
mechanism	O
to	O
a	O
desired	O
payment	O
bound	O
.	O
The	O
ratio	O
captures	O
the	O
extent	O
to	O
which	O
the	O
mechanism	O
overpays	O
,	O
relative	O
to	O
perceived	O
fair	O
cost	O
in	O
a	O
truthful	O
auction	B-KEY
.	O
In	O
this	O
paper	O
,	O
we	O
propose	O
a	O
new	O
truthful	O
polynomial-time	O
auction	B-KEY
for	O
the	O
vertex	B-KEY
cover	I-KEY
problem	O
and	O
bound	O
its	O
frugality	B-KEY
ratio	O
.	O
We	O
show	O
that	O
the	O
solution	O
quality	O
is	O
with	O
a	O
constant	O
factor	O
of	O
optimal	O
and	O
the	O
frugality	B-KEY
ratio	O
is	O
within	O
a	O
constant	O
factor	O
of	O
the	O
best	O
possible	O
worst-case	O
bound	O
;	O
this	O
is	O
the	O
first	O
auction	O
for	O
this	O
problem	O
to	O
have	O
these	O
properties	O
.	O
Moreover	O
,	O
we	O
show	O
how	O
to	O
transform	O
any	O
truthful	O
auction	B-KEY
into	O
a	O
frugal	B-KEY
one	O
while	O
preserving	O
the	O
approximation	O
ratio	O
.	O
Also	O
,	O
we	O
consider	O
two	O
natural	O
modifications	O
of	O
the	O
definition	O
of	O
Karlin	O
et	O
al.	O
,	O
and	O
we	O
analyse	O
the	O
properties	O
of	O
the	O
resulting	O
payment	O
bounds	O
,	O
such	O
as	O
monotonicity	O
,	O
computational	O
hardness	O
,	O
and	O
robustness	O
with	O
respect	O
to	O
the	O
draw-resolution	O
rule	O
.	O
We	O
study	O
the	O
relationships	O
between	O
the	O
different	O
payment	O
bounds	O
,	O
both	O
for	O
general	O
set	O
systems	O
and	O
for	O
specific	O
set-system	O
auctions	B-KEY
,	O
such	O
as	O
path	O
auctions	B-KEY
and	O
vertex-cover	O
auctions	B-KEY
.	O
We	O
use	O
these	O
new	O
definitions	O
in	O
the	O
proof	O
of	O
our	O
main	O
result	O
for	O
vertex-cover	O
auctions	B-KEY
via	O
a	O
bootstrapping	B-KEY
technique	I-KEY
,	O
which	O
may	O
be	O
of	O
independent	O
interest	O
.	O

Information	B-KEY
Searching	I-KEY
and	I-KEY
Sharing	I-KEY
in	O
Large-Scale	O
Dynamic	O
Networks	O
ABSTRACT	O
Finding	O
the	O
right	O
agents	O
in	O
a	O
large	O
and	O
dynamic	O
network	O
to	O
provide	O
the	O
needed	O
resources	O
in	O
a	O
timely	O
fashion	O
,	O
is	O
a	O
long	O
standing	O
problem	O
.	O
This	O
paper	O
presents	O
a	O
method	O
for	O
information	B-KEY
searching	I-KEY
and	I-KEY
sharing	I-KEY
that	O
combines	O
routing	O
indices	O
with	O
tokenbased	O
methods	O
.	O
The	O
proposed	O
method	O
enables	O
agents	O
to	O
search	O
effectively	O
by	O
acquiring	O
their	O
neighbors	O
'	O
interests	O
,	O
advertising	O
their	O
information	O
provision	O
abilities	O
and	O
maintaining	O
indices	O
for	O
routing	O
queries	O
,	O
in	O
an	O
integrated	O
way	O
.	O
Specifically	O
,	O
the	O
paper	O
demonstrates	O
through	O
performance	B-KEY
experiments	O
how	O
static	O
and	O
dynamic	O
networks	O
of	O
agents	O
can	O
be	O
`	O
tuned	O
'	O
to	O
answer	O
queries	O
effectively	O
as	O
they	O
gather	O
evidence	O
for	O
the	O
interests	O
and	O
information	O
provision	O
abilities	O
of	O
others	O
,	O
without	O
altering	O
the	O
topology	O
or	O
imposing	O
an	O
overlay	O
structure	O
to	O
the	O
network	O
of	O
acquaintances	O
.	O

Trading	B-KEY
Networks	I-KEY
with	O
Price-Setting	O
Agents	O
ABSTRACT	O
In	O
a	O
wide	O
range	O
of	O
markets	B-KEY
,	O
individual	O
buyers	O
and	O
sellers	O
often	O
trade	O
through	O
intermediaries	O
,	O
who	O
determine	O
prices	O
via	O
strategic	O
considerations	O
.	O
Typically	O
,	O
not	O
all	O
buyers	O
and	O
sellers	O
have	O
access	O
to	O
the	O
same	O
intermediaries	O
,	O
and	O
they	O
trade	O
at	O
correspondingly	O
different	O
prices	O
that	O
reflect	O
their	O
relative	O
amounts	O
of	O
power	O
in	O
the	O
market	B-KEY
.	O
We	O
model	O
this	O
phenomenon	O
using	O
a	O
game	O
in	O
which	O
buyers	O
,	O
sellers	O
,	O
and	O
traders	O
engage	O
in	O
trade	O
on	O
a	O
graph	O
that	O
represents	O
the	O
access	O
each	O
buyer	O
and	O
seller	O
has	O
to	O
the	O
traders	O
.	O
In	O
this	O
model	O
,	O
traders	O
set	O
prices	O
strategically	O
,	O
and	O
then	O
buyers	O
and	O
sellers	O
react	O
to	O
the	O
prices	O
they	O
are	O
offered	O
.	O
We	O
show	O
that	O
the	O
resulting	O
game	O
always	O
has	O
a	O
subgame	O
perfect	O
Nash	O
equilibrium	O
,	O
and	O
that	O
all	O
equilibria	O
lead	O
to	O
an	O
efficient	O
-LRB-	O
i.e.	O
socially	O
optimal	O
-RRB-	O
allocation	O
of	O
goods	O
.	O
We	O
extend	O
these	O
results	O
to	O
a	O
more	O
general	O
type	O
of	O
matching	O
market	B-KEY
,	O
such	O
as	O
one	O
finds	O
in	O
the	O
matching	O
of	O
job	O
applicants	O
and	O
employers	O
.	O
Finally	O
,	O
we	O
consider	O
how	O
the	O
profits	O
obtained	O
by	O
the	O
traders	O
depend	O
on	O
the	O
underlying	O
graph	O
--	O
roughly	O
,	O
a	O
trader	O
can	O
command	O
a	O
positive	O
profit	O
if	O
and	O
only	O
if	O
it	O
has	O
an	O
``	O
essential	O
''	O
connection	O
in	O
the	O
network	O
structure	O
,	O
thus	O
providing	O
a	O
graph-theoretic	O
basis	O
for	O
quantifying	O
the	O
amount	O
of	O
competition	O
among	O
traders	O
.	O
Our	O
work	O
differs	O
from	O
recent	O
studies	O
of	O
how	O
price	O
is	O
affected	O
by	O
network	O
structure	O
through	O
our	O
modeling	O
of	O
price-setting	O
as	O
a	O
strategic	O
activity	O
carried	O
out	O
by	O
a	O
subset	O
of	O
agents	O
in	O
the	O
system	O
,	O
rather	O
than	O
studying	O
prices	O
set	O
via	O
competitive	O
equilibrium	O
or	O
by	O
a	O
truthful	O
mechanism	O
.	O

An	O
Initial	O
Analysis	O
and	O
Presentation	O
of	O
Malware	B-KEY
Exhibiting	O
Swarm-Like	O
Behavior	O
ABSTRACT	O
The	O
Slammer	O
,	O
which	O
is	O
currently	O
the	O
fastest	O
computer	O
worm	O
in	O
recorded	O
history	O
,	O
was	O
observed	O
to	O
infect	O
90	O
percent	O
of	O
all	O
vulnerable	O
Internets	O
hosts	O
within	O
10	O
minutes	O
.	O
Although	O
the	O
main	O
action	O
that	O
the	O
Slammer	B-KEY
worm	I-KEY
takes	O
is	O
a	O
relatively	O
unsophisticated	O
replication	O
of	O
itself	O
,	O
it	O
still	O
spreads	O
so	O
quickly	O
that	O
human	O
response	O
was	O
ineffective	O
.	O
Most	O
proposed	O
countermeasures	O
strategies	O
are	O
based	O
primarily	O
on	O
rate	O
detection	O
and	O
limiting	O
algorithms	O
.	O
However	O
,	O
such	O
strategies	O
are	O
being	O
designed	O
and	O
developed	O
to	O
effectively	O
contain	O
worms	O
whose	O
behaviors	O
are	O
similar	O
to	O
that	O
of	O
Slammer	O
.	O
In	O
our	O
work	O
,	O
we	O
put	O
forth	O
the	O
hypothesis	O
that	O
next	O
generation	O
worms	O
will	O
be	O
radically	O
different	O
,	O
and	O
potentially	O
such	O
techniques	O
will	O
prove	O
ineffective	O
.	O
Specifically	O
,	O
we	O
propose	O
to	O
study	O
a	O
new	O
generation	O
of	O
worms	O
called	O
''	O
Swarm	B-KEY
Worms	I-KEY
''	O
,	O
whose	O
behavior	O
is	O
predicated	O
on	O
the	O
concept	O
of	O
''	O
emergent	B-KEY
intelligence	I-KEY
''	O
.	O
Emergent	B-KEY
Intelligence	I-KEY
is	O
the	O
behavior	O
of	O
systems	O
,	O
very	O
much	O
like	O
biological	O
systems	O
such	O
as	O
ants	O
or	O
bees	O
,	O
where	O
simple	O
local	O
interactions	O
of	O
autonomous	O
members	O
,	O
with	O
simple	O
primitive	O
actions	O
,	O
gives	O
rise	O
to	O
complex	O
and	O
intelligent	O
global	O
behavior	O
.	O
In	O
this	O
manuscript	O
we	O
will	O
introduce	O
the	O
basic	O
principles	O
behind	O
the	O
idea	O
of	O
''	O
Swarm	B-KEY
Worms	I-KEY
''	O
,	O
as	O
well	O
as	O
the	O
basic	O
structure	O
required	O
in	O
order	O
to	O
be	O
considered	O
a	O
''	O
swarm	B-KEY
worm	I-KEY
''	O
.	O
In	O
addition	O
,	O
we	O
will	O
present	O
preliminary	O
results	O
on	O
the	O
propagation	O
speeds	O
of	O
one	O
such	O
swarm	B-KEY
worm	I-KEY
,	O
called	O
the	O
ZachiK	B-KEY
worm	O
.	O
We	O
will	O
show	O
that	O
ZachiK	B-KEY
is	O
capable	O
of	O
propagating	O
at	O
a	O
rate	O
2	O
orders	O
of	O
magnitude	O
faster	O
than	O
similar	O
worms	O
without	O
swarm	O
capabilities	O
.	O

Betting	O
on	O
Permutations	O
ABSTRACT	O
We	O
consider	O
a	O
permutation	B-KEY
betting	I-KEY
scenario	O
,	O
where	O
people	O
wager	O
on	O
the	O
final	O
ordering	O
of	O
n	O
candidates	O
:	O
for	O
example	O
,	O
the	O
outcome	O
of	O
a	O
horse	O
race	O
.	O
We	O
examine	O
the	O
auctioneer	O
problem	O
of	O
risklessly	O
matching	O
up	O
wagers	O
or	O
,	O
equivalently	O
,	O
finding	O
arbitrage	O
opportunities	O
among	O
the	O
proposed	O
wagers	O
.	O
Requiring	O
bidders	O
to	O
explicitly	O
list	O
the	O
orderings	O
that	O
they	O
'd	O
like	O
to	O
bet	O
on	O
is	O
both	O
unnatural	O
and	O
intractable	O
,	O
because	O
the	O
number	O
of	O
orderings	O
is	O
n	O
!	O
and	O
the	O
number	O
of	O
subsets	O
of	O
orderings	O
is	O
2n	O
!	O
.	O
We	O
propose	O
two	O
expressive	B-KEY
betting	I-KEY
languages	O
that	O
seem	O
natural	O
for	O
bidders	O
,	O
and	O
examine	O
the	O
computational	B-KEY
complexity	I-KEY
of	O
the	O
auctioneer	O
problem	O
in	O
each	O
case	O
.	O
Subset	B-KEY
betting	I-KEY
allows	O
traders	O
to	O
bet	O
either	O
that	O
a	O
candidate	O
will	O
end	O
up	O
ranked	O
among	O
some	O
subset	O
of	O
positions	O
in	O
the	O
final	O
ordering	O
,	O
for	O
example	O
,	O
``	O
horse	O
A	O
will	O
finish	O
in	O
positions	O
4	O
,	O
9	O
,	O
or	O
13-21	O
''	O
,	O
or	O
that	O
a	O
position	O
will	O
be	O
taken	O
by	O
some	O
subset	O
of	O
candidates	O
,	O
for	O
example	O
``	O
horse	O
A	O
,	O
B	O
,	O
or	O
D	O
will	O
finish	O
in	O
position	O
2	O
''	O
.	O
For	O
subset	B-KEY
betting	I-KEY
,	O
we	O
show	O
that	O
the	O
auctioneer	O
problem	O
can	O
be	O
solved	O
in	O
polynomial	O
time	O
if	O
orders	O
are	O
divisible	O
.	O
Pair	O
betting	O
allows	O
traders	O
to	O
bet	O
on	O
whether	O
one	O
candidate	O
will	O
end	O
up	O
ranked	O
higher	O
than	O
another	O
candidate	O
,	O
for	O
example	O
``	O
horse	O
A	O
will	O
beat	O
horse	O
B	O
''	O
.	O
We	O
prove	O
that	O
the	O
auctioneer	O
problem	O
becomes	O
NP-hard	O
for	O
pair	O
betting	O
.	O
We	O
identify	O
a	O
sufficient	O
condition	O
for	O
the	O
existence	O
of	O
a	O
pair	O
betting	O
match	O
that	O
can	O
be	O
verified	O
in	O
polynomial	O
time	O
.	O
We	O
also	O
show	O
that	O
a	O
natural	O
greedy	B-KEY
algorithm	I-KEY
gives	O
a	O
poor	O
approximation	O
for	O
indivisible	O
orders	O
.	O

Design	O
and	O
Implementation	O
of	O
a	O
Distributed	B-KEY
Content	I-KEY
Management	I-KEY
System	O
ABSTRACT	O
The	O
convergence	O
of	O
advances	O
in	O
storage	O
,	O
encoding	O
,	O
and	O
networking	O
technologies	O
has	O
brought	O
us	O
to	O
an	O
environment	O
where	O
huge	O
amounts	O
of	O
continuous	O
media	O
content	O
is	O
routinely	O
stored	O
and	O
exchanged	O
between	O
network	O
enabled	O
devices	O
.	O
Keeping	O
track	O
of	O
-LRB-	O
or	O
managing	O
-RRB-	O
such	O
content	O
remains	O
challenging	O
due	O
to	O
the	O
sheer	O
volume	O
of	O
data	O
.	O
Storing	O
``	O
live	O
''	O
continuous	O
media	O
-LRB-	O
such	O
as	O
TV	O
or	O
radio	O
content	O
-RRB-	O
adds	O
to	O
the	O
complexity	O
in	O
that	O
this	O
content	O
has	O
no	O
well	O
defined	O
start	O
or	O
end	O
and	O
is	O
therefore	O
cumbersome	O
to	O
deal	O
with	O
.	O
Networked	O
storage	O
allows	O
content	O
that	O
is	O
logically	O
viewed	O
as	O
part	O
of	O
the	O
same	O
collection	O
to	O
in	O
fact	O
be	O
distributed	O
across	O
a	O
network	O
,	O
making	O
the	O
task	O
of	O
content	O
management	O
all	O
but	O
impossible	O
to	O
deal	O
with	O
without	O
a	O
content	O
management	O
system	O
.	O
In	O
this	O
paper	O
we	O
present	O
the	O
design	O
and	O
implementation	O
of	O
the	O
Spectrum	B-KEY
content	I-KEY
management	I-KEY
system	I-KEY
,	O
which	O
deals	O
with	O
rich	O
media	O
content	O
effectively	O
in	O
this	O
environment	O
.	O
Spectrum	O
has	O
a	O
modular	O
architecture	O
that	O
allows	O
its	O
application	O
to	O
both	O
stand-alone	O
and	O
various	O
networked	O
scenarios	O
.	O
A	O
unique	O
aspect	O
of	O
Spectrum	O
is	O
that	O
it	O
requires	O
one	O
-LRB-	O
or	O
more	O
-RRB-	O
retention	O
policies	O
to	O
apply	O
to	O
every	O
piece	O
of	O
content	O
that	O
is	O
stored	O
in	O
the	O
system	O
.	O
This	O
means	O
that	O
there	O
are	O
no	O
eviction	O
policies	O
.	O
Content	O
that	O
no	O
longer	O
has	O
a	O
retention	O
policy	O
applied	O
to	O
it	O
is	O
simply	O
removed	O
from	O
the	O
system	O
.	O
Different	O
retention	O
policies	O
can	O
easily	O
be	O
applied	O
to	O
the	O
same	O
content	O
thus	O
naturally	O
facilitating	O
sharing	O
without	O
duplication	O
.	O
This	O
approach	O
also	O
allows	O
Spectrum	O
to	O
easily	O
apply	O
time	O
based	O
policies	O
which	O
are	O
basic	O
building	O
blocks	O
required	O
to	O
deal	O
with	O
the	O
storage	O
of	O
live	O
continuous	O
media	O
,	O
to	O
content	O
.	O
We	O
not	O
only	O
describe	O
the	O
details	O
of	O
the	O
Spectrum	O
architecture	O
but	O
also	O
give	O
typical	O
use	O
cases	O
.	O

Latent	O
Concept	O
Expansion	O
Using	O
Markov	B-KEY
Random	I-KEY
Fields	I-KEY
ABSTRACT	O
Query	B-KEY
expansion	I-KEY
,	O
in	O
the	O
form	O
of	O
pseudo-relevance	B-KEY
feedback	I-KEY
or	O
relevance	O
feedback	O
,	O
is	O
a	O
common	O
technique	O
used	O
to	O
improve	O
retrieval	O
effectiveness	O
.	O
Most	O
previous	O
approaches	O
have	O
ignored	O
important	O
issues	O
,	O
such	O
as	O
the	O
role	O
of	O
features	O
and	O
the	O
importance	O
of	O
modeling	O
term	O
dependencies	O
.	O
In	O
this	O
paper	O
,	O
we	O
propose	O
a	O
robust	O
query	B-KEY
expansion	I-KEY
technique	O
based	O
on	O
the	O
Markov	O
random	O
field	O
model	O
for	O
information	O
retrieval	O
.	O
The	O
technique	O
,	O
called	O
latent	O
concept	O
expansion	O
,	O
provides	O
a	O
mechanism	O
for	O
modeling	O
term	O
dependencies	O
during	O
expansion	O
.	O
Furthermore	O
,	O
the	O
use	O
of	O
arbitrary	O
features	O
within	O
the	O
model	O
provides	O
a	O
powerful	O
framework	O
for	O
going	O
beyond	O
simple	O
term	O
occurrence	O
features	O
that	O
are	O
implicitly	O
used	O
by	O
most	O
other	O
expansion	O
techniques	O
.	O
We	O
evaluate	O
our	O
technique	O
against	O
relevance	O
models	O
,	O
a	O
state-of-the-art	O
language	O
modeling	O
query	B-KEY
expansion	I-KEY
technique	O
.	O
Our	O
model	O
demonstrates	O
consistent	O
and	O
significant	O
improvements	O
in	O
retrieval	O
effectiveness	O
across	O
several	O
TREC	O
data	O
sets	O
.	O
We	O
also	O
describe	O
how	O
our	O
technique	O
can	O
be	O
used	O
to	O
generate	O
meaningful	O
multi-term	O
concepts	O
for	O
tasks	O
such	O
as	O
query	O
suggestion/reformulation	O
.	O

Meta-Level	O
Coordination	O
for	O
Solving	O
Negotiation	B-KEY
Chains	I-KEY
in	O
Semi-Cooperative	O
Multi-Agent	O
Systems	O
ABSTRACT	O
A	O
negotiation	B-KEY
chain	I-KEY
is	O
formed	O
when	O
multiple	O
related	O
negotiations	O
are	O
spread	O
over	O
multiple	B-KEY
agents	I-KEY
.	O
In	O
order	O
to	O
appropriately	O
order	O
and	O
structure	O
the	O
negotiations	O
occurring	O
in	O
the	O
chain	O
so	O
as	O
to	O
optimize	O
the	O
expected	O
utility	O
,	O
we	O
present	O
an	O
extension	O
to	O
a	O
singleagent	O
concurrent	O
negotiation	B-KEY
framework	I-KEY
.	O
This	O
work	O
is	O
aimed	O
at	O
semi-cooperative	O
multi-agent	O
systems	O
,	O
where	O
each	O
agent	B-KEY
has	O
its	O
own	O
goals	O
and	O
works	O
to	O
maximize	O
its	O
local	O
utility	O
;	O
however	O
,	O
the	O
performance	O
of	O
each	O
individual	O
agent	B-KEY
is	O
tightly	O
related	O
to	O
other	O
agent	B-KEY
's	O
cooperation	O
and	O
the	O
system	O
's	O
overall	O
performance	O
.	O
We	O
introduce	O
a	O
pre-negotiation	B-KEY
phase	O
that	O
allows	O
agents	B-KEY
to	O
transfer	O
meta-level	O
information	O
.	O
Using	O
this	O
information	O
,	O
the	O
agent	B-KEY
can	O
build	O
a	O
more	O
accurate	O
model	O
of	O
the	O
negotiation	O
in	O
terms	O
of	O
modeling	O
the	O
relationship	O
of	O
flexibility	B-KEY
and	O
success	O
probability	O
.	O
This	O
more	O
accurate	O
model	O
helps	O
the	O
agent	B-KEY
in	O
choosing	O
a	O
better	O
negotiation	O
solution	O
in	O
the	O
global	O
negotiation	B-KEY
chain	I-KEY
context	O
.	O
The	O
agent	B-KEY
can	O
also	O
use	O
this	O
information	O
to	O
allocate	O
appropriate	O
time	O
for	O
each	O
negotiation	O
,	O
hence	O
to	O
find	O
a	O
good	O
ordering	O
of	O
all	O
related	O
negotiations	O
.	O
The	O
experimental	O
data	O
shows	O
that	O
these	O
mechanisms	O
improve	O
the	O
agents	B-KEY
'	O
and	O
the	O
system	O
's	O
overall	O
performance	O
significantly	O
.	O

The	O
Influence	O
of	O
Caption	B-KEY
Features	I-KEY
on	O
Clickthrough	B-KEY
Patterns	I-KEY
in	O
Web	B-KEY
Search	I-KEY
ABSTRACT	O
Web	B-KEY
search	I-KEY
engines	O
present	O
lists	O
of	O
captions	O
,	O
comprising	O
title	O
,	O
snippet	B-KEY
,	O
and	O
URL	O
,	O
to	O
help	O
users	O
decide	O
which	O
search	O
results	O
to	O
visit	O
.	O
Understanding	O
the	O
influence	O
of	O
features	O
of	O
these	O
captions	O
on	O
Web	B-KEY
search	I-KEY
behavior	O
may	O
help	O
validate	O
algorithms	O
and	O
guidelines	O
for	O
their	O
improved	O
generation	O
.	O
In	O
this	O
paper	O
we	O
develop	O
a	O
methodology	O
to	O
use	O
clickthrough	O
logs	O
from	O
a	O
commercial	O
search	O
engine	O
to	O
study	O
user	O
behavior	O
when	O
interacting	O
with	O
search	O
result	O
captions	O
.	O
The	O
findings	O
of	O
our	O
study	O
suggest	O
that	O
relatively	O
simple	O
caption	B-KEY
features	I-KEY
such	O
as	O
the	O
presence	O
of	O
all	O
terms	O
query	O
terms	O
,	O
the	O
readability	O
of	O
the	O
snippet	B-KEY
,	O
and	O
the	O
length	O
of	O
the	O
URL	O
shown	O
in	O
the	O
caption	O
,	O
can	O
significantly	O
influence	O
users	O
'	O
Web	B-KEY
search	I-KEY
behavior	O
.	O

Utility-based	O
Information	O
Distillation	O
Over	O
Temporally	O
Sequenced	O
Documents	O
ABSTRACT	O
This	O
paper	O
examines	O
a	O
new	O
approach	O
to	O
information	O
distillation	O
over	O
temporally	B-KEY
ordered	I-KEY
documents	I-KEY
,	O
and	O
proposes	O
a	O
novel	O
evaluation	O
scheme	O
for	O
such	O
a	O
framework	O
.	O
It	O
combines	O
the	O
strengths	O
of	O
and	O
extends	O
beyond	O
conventional	O
adaptive	B-KEY
filtering	I-KEY
,	O
novelty	B-KEY
detection	I-KEY
and	O
non-redundant	O
passage	B-KEY
ranking	I-KEY
with	O
respect	O
to	O
long-lasting	O
information	O
needs	O
-LRB-	O
`	O
tasks	O
'	O
with	O
multiple	O
queries	O
-RRB-	O
.	O
Our	O
approach	O
supports	O
fine-grained	O
user	O
feedback	O
via	O
highlighting	O
of	O
arbitrary	O
spans	O
of	O
text	O
,	O
and	O
leverages	O
such	O
information	O
for	O
utility	O
optimization	O
in	O
adaptive	O
settings	O
.	O
For	O
our	O
experiments	O
,	O
we	O
defined	O
hypothetical	O
tasks	O
based	O
on	O
news	O
events	O
in	O
the	O
TDT4	O
corpus	O
,	O
with	O
multiple	O
queries	O
per	O
task	O
.	O
Answer	O
keys	O
-LRB-	O
nuggets	O
-RRB-	O
were	O
generated	O
for	O
each	O
query	O
and	O
a	O
semiautomatic	O
procedure	O
was	O
used	O
for	O
acquiring	O
rules	O
that	O
allow	O
automatically	O
matching	O
nuggets	O
against	O
system	O
responses	O
.	O
We	O
also	O
propose	O
an	O
extension	O
of	O
the	O
NDCG	B-KEY
metric	I-KEY
for	O
assessing	O
the	O
utility	O
of	O
ranked	O
passages	O
as	O
a	O
combination	O
of	O
relevance	O
and	O
novelty	O
.	O
Our	O
results	O
show	O
encouraging	O
utility	O
enhancements	O
using	O
the	O
new	O
approach	O
,	O
compared	O
to	O
the	O
baseline	O
systems	O
without	O
incremental	O
learning	O
or	O
the	O
novelty	B-KEY
detection	I-KEY
components	O
.	O

Operation	B-KEY
Context	I-KEY
and	O
Context-based	O
Operational	B-KEY
Transformation	I-KEY
ABSTRACT	O
Operational	B-KEY
Transformation	I-KEY
-LRB-	O
OT	B-KEY
-RRB-	O
is	O
a	O
technique	O
for	O
consistency	B-KEY
maintenance	I-KEY
and	O
group	O
undo	B-KEY
,	O
and	O
is	O
being	O
applied	O
to	O
an	O
increasing	O
number	O
of	O
collaborative	O
applications	O
.	O
The	O
theoretical	O
foundation	O
for	O
OT	B-KEY
is	O
crucial	O
in	O
determining	O
its	O
capability	O
to	O
solve	O
existing	O
and	O
new	O
problems	O
,	O
as	O
well	O
as	O
the	O
quality	O
of	O
those	O
solutions	O
.	O
The	O
theory	O
of	O
causality	O
has	O
been	O
the	O
foundation	O
of	O
all	O
prior	O
OT	B-KEY
systems	O
,	O
but	O
it	O
is	O
inadequate	O
to	O
capture	O
essential	O
correctness	O
requirements	O
.	O
Past	O
research	O
had	O
invented	O
various	O
patches	O
to	O
work	O
around	O
this	O
problem	O
,	O
resulting	O
in	O
increasingly	O
intricate	O
and	O
complicated	O
OT	B-KEY
algorithms	O
.	O
After	O
having	O
designed	O
,	O
implemented	O
,	O
and	O
experimented	O
with	O
a	O
series	O
of	O
OT	B-KEY
algorithms	O
,	O
we	O
reflected	O
on	O
what	O
had	O
been	O
learned	O
and	O
set	O
out	O
to	O
develop	O
a	O
new	O
theoretical	O
framework	O
for	O
better	O
understanding	O
and	O
resolving	O
OT	B-KEY
problems	O
,	O
reducing	O
its	O
complexity	O
,	O
and	O
supporting	O
its	O
continual	O
evolution	O
.	O
In	O
this	O
paper	O
,	O
we	O
report	O
the	O
main	O
results	O
of	O
this	O
effort	O
:	O
the	O
theory	O
of	O
operation	B-KEY
context	I-KEY
and	O
the	O
COT	B-KEY
-LRB-	O
Context-based	O
OT	B-KEY
-RRB-	O
algorithm	O
.	O
The	O
COT	B-KEY
algorithm	O
is	O
capable	O
of	O
supporting	O
both	O
do	O
and	O
undo	B-KEY
of	O
any	O
operations	O
at	O
anytime	O
,	O
without	O
requiring	O
transformation	O
functions	O
to	O
preserve	O
Reversibility	O
Property	O
,	O
Convergence	O
Property	O
2	O
,	O
Inverse	O
Properties	O
2	O
and	O
3	O
.	O
The	O
COT	B-KEY
algorithm	O
is	O
not	O
only	O
simpler	O
and	O
more	O
efficient	O
than	O
prior	O
OT	B-KEY
control	O
algorithms	O
,	O
but	O
also	O
simplifies	O
the	O
design	O
of	O
transformation	O
functions	O
.	O
We	O
have	O
implemented	O
the	O
COT	B-KEY
algorithm	O
in	O
a	O
generic	O
collaboration	O
engine	O
and	O
used	O
it	O
for	O
supporting	O
a	O
range	O
of	O
novel	O
collaborative	O
applications	O
.	O