Datasets:
ulab-ai
/
arxiv-tables

Dataset card Data Studio Files
xet
Community
id
int64
1
213k
paper_arxiv_id
stringlengths
10
13
path
float64
caption
stringlengths
9
127k
⌀
label
stringlengths
7
125
⌀
table_text
stringlengths
33
131k
⌀
212,602
2305.01034v2
null
\caption{Mapping different learning settings under a common set of notation.}
\label{tab:mapping}
\begin{tabular}{llll} \toprule {\bf Setting} & $x$ & {\bf Features} & $f^*(x)$ \\ \midrule {\bf Supervised classification} & input & $d_1:$ class & class \\ & & $c_1:$ instance within class & \\ \\ {\bf Reinforcement learning} & state or observation & $c_1:$ state or & desired action \\ & sequence & observation sequence & \\ \\ {\bf Meta-learning for} & samples from a set & $c_1$: a set of related classes & classifier mapping inputs \\ {\bf few-shot classification} & of related classes & $c_2$: choice of samples from $c_1$ & to their class \\ \\ {\bf Unsupervised autoencoding} & input & $c_1$: input & $x$ \\ & & & \\ \\ {\bf Meta-reinforcement learning} & trajectories through & $c_1$: environment & policy mapping states \\ & environments & $c_2$: trajectory within environment & to desired actions \\\bottomrule \end{tabular}
212,603
2305.01034v2
null
\caption{The inductive bias information contributed by different model architectures for image classification tasks. (FC-$N$ refers to a fully connected network with $N$ layers.)}
\label{tab:arch_full}
\begin{tabular}{p{0.45\textwidth}|p{0.20\textwidth}} \toprule \multicolumn{2}{c}{\textbf{ImageNet}} \\ \hline Model & Information Content ({\small$\times 10^{41}$ bits}) \\ \hhline{=|=} Linear \citep{karpathy2015breaking} & $2.063$\\ \hline SIFT + FVs \citep{sanchez2011high} & $2.261$ \\ \hline AlexNet \citep{krizhevsky2012imagenet} & $2.290$ \\ \hline DenseNet-121 \citep{huang2017densely} & $2.348$ \\ \hline ResNet-50 \citep{he2016resnet} & $2.362$ \\ \hline DenseNet-201 \citep{huang2017densely} & $2.363$ \\ \hline \small{WRN-50-2-bottleneck} \citep{zagoruyko2016wide} & $2.368$ \\ \hline BiT-L \citep{kolesnikov2020bit} & $2.449$ \\ \hline ViT-H/14 \citep{dosovitskiy2021image} & $2.461$ \\ \hline \end{tabular}