TexOCR
Collection
7 items • Updated
latex_text string | figure_path string |
|---|---|
\begin{figure}
\centering
% Figure 5
\includegraphics[width=\textwidth]{figure_5.pdf}
\caption{The HOSIDF-based validation of the designed control; (a) Open loop, (b) Sensitivity function, (c) Complementary sensitivity function.}
\label{figure_5}
\end{figure}
\begin{figure}
\centering
% Figure 6
\includegraphics[width... | 2510.09445v1_5.png |
level performance measure or Level of Service (LoS). The LoS aggregates individual assets’ conditions into a single representative indicator; for instance, a weighted sum of pavement quality indices for different roads. Maximizing the long-term LoS thus becomes the principal objective, although other objectives—such as... | 2507.18732v1_6.png |
\section{References}
@article{Bai_2023,
author = {Bai, S. and Chen, K. and Liu, X. and Wang, J. and Ge, W. and Song, S. and Dang, K. and Wang, Y. and Wang, S. and Tang, J. and Zhong, H. and Zhu, Y. and Yang, M. and Li, Z. and Wan, J. and Wang, P. and Ding, W. and Fu, Z. and Xu, Y. and Ye, J. and Zhang, X. and Xie, T... | 2509.18711v1_7.png |
\begin{figure}[t]
\centering
\includegraphics[width=\linewidth]{figure_1.pdf}
\caption{Example of how redescriptions can be expressed in the framework of Problem~\ref{prob:exact}. \textbf{Top:} A pair of queries, forming a redescription. \textbf{Bottom left and middle:} Original data matrices $\mD_L$ and $\mD_R$ for re... | 2507.08745v1_4.png |
\begin{figure}[t]
\centering
\includegraphics[width=\linewidth]{figure_8.pdf}
\caption{Energy consumption (in Joules) for the hardware-constrained baselines of MUSIC (in black), ESPRIT (in red), and Root-MUSIC (in gray), and the adaptive MUSIC case (in blue), per power mode across the vehicle's trajectory.}
\label{figu... | 2507.20399v1_18.png |
at the active site. This balance is essential for industrial enzymes operating under thermal stress, as excessive rigidity can impair catalytic efficiency.
While these results collectively highlight segment transformer’s ability to guide thermostability-related enzyme engineering, experimental data from a single enzym... | 2507.19755v1_25.png |
\title{Two-Stage TSO-DSO Services Provision Framework for Electric Vehicle Coordination}
\author{Yi Wang, Member, IEEE, Dawei Qiu, Member, IEEE, Fei Teng, Senior Member, IEEE, and Goran Strbac, Member, IEEE}
\begin{abstract}
High renewable penetration has been witnessed in power systems, resulting in reduced system i... | 2507.18110v1_0.png |
\begin{figure}[t]
\centering
\includegraphics[width=\linewidth]{figure_1.pdf}
\caption{Energy per mole for the topochemical reactions between stoichiometric LiBC and metal salts XM (X = I, Br, Cl, or NO$_3$) resulting in the formation of MBC (M = Cu or Ag). Negative energies indicate exothermic reactions.}
\label{figur... | 2507.14281v2_2.png |
where $b(s)$ is a baseline (often a state-value function) to reduce variance in the gradient estimate. Algorithms like REINFORCE or actor-critic variants (e.g., A2C, PPO) use Monte Carlo or bootstrapping techniques to estimate $G_t$ and update $\boldsymbol{\theta}$ accordingly. In contrast to value-based methods that d... | 2507.19458v1_16.png |
\title{Queue up for Takeoff: A Transferable Deep Learning Framework for Flight Delay Prediction}
\author{Nnamdi Daniel Aghanya, Ta Duong Vu\\
Cranfield University, Cranfield\\
\texttt{\{nnamdi.aghanya, duong.vu\}@cranfield.ac.uk}
\and
Amaëlle Diop, Charlotte Deville, Nour Imane Kerroumi\\
Cranfield University, Cranfie... | 2507.09084v1_0.png |
types tend to specialize in particular biological domains, with limited generalizability beyond their source modality. Among them, literature-derived gene embeddings and language models (LMs) showed relatively strong and consistent performance across a broad range of benchmarks, highlighting their potential as versatil... | 2507.07367v1_1.png |
\begin{figure}[t]
\centering
\includegraphics[width=\linewidth]{figure_1.pdf}
\caption{(a) The difference between attribute-driven multi-interest analysis and LLM-driven multi-interest analysis. (b) LLM-driven multi-interest analysis across varying levels of granularity. (c) The sparsity of the user's behaviors compare... | 2507.10917v2_1.png |
% Acknowledgements section
\section*{Acknowledgements}
This work is funded by the European Union Horizon 2020 research and innovation programme under the Marie Sk\l{}odowska--Curie grant agreement No 860801 and the FWF Austrian Science Fund [10.55776/COE12]. Sabrina Kirrane is also funded by the FWF Austrian Science F... | 2510.04652v1_19.png |
\title{Comprehension Without Competence: Architectural Limits of LLMs in Symbolic Computation and Reasoning*}
\author{Zheng Zhang\\Amazon Web Service\\zha@amazon.com, zzhang@gmail.com}
\begin{abstract}
Large Language Models (LLMs) display striking surface fluency yet systematically fail at tasks requiring symbolic re... | 2507.10624v1_0.png |
\begin{table*}[t]
\centering
\scalebox{0.8}{
\begin{tabular}{l|l|cccccc|ccc|c}
\toprule
Model & Year & BLEU-1 & BLEU-2 & BLEU-3 & BLEU-4 & METEOR & ROUGE & Precision & Recall & F1 & Avg \\
\midrule
R2Gen & ACL 2020 & 0.289 & 0.155 & 0.087 & 0.052 & 0.128 & 0.243 & 0.151 & 0.145 & 0.145 & 0.155 \\
M2... | 2507.07568v1_6.png |
\begin{figure}[t]
\centering
\includegraphics[width=\linewidth]{figure_4.pdf}
\caption{P95 latency comparison across traffic patterns for KIScaler, GPU baseline, and CPU baseline.}
\label{figure_4}
\end{figure}
\begin{table}[htbp]
\centering
\caption{P95 Latency Comparison: KIScaler vs. Baselines}
\label{table_4}
\res... | 2507.07932v1_6.png |
One of the most important randomized tools for computational linear algebra is matrix sketching, along with its underlying theoretical framework of subspace embeddings \cite{53}. In the context of solving linear systems, sketching-based techniques have focused primarily on the highly over-determined regression setting ... | 2507.11724v1_5.png |
\begin{table}[h!]
\centering
\caption{Cell capacity measurements (in amperhour) on three modules (M1, M2, M3). Data organised by cell position inside each module (H1--6, L1--6)}
\begin{tabular}{c|ccc||c|ccc}
& M1 & M2 & M3 & & M1 & M2 & M3 \\
\hline
H1 & 93.8 & 91.6 & 91.0 & L1 & 93.4 & 88.4 & 89.5 \\
H2 & 92.6 & 88.... | 2507.14020v1_2.png |
Note that $\widetilde{U}(\boldsymbol{\beta}, \boldsymbol{\gamma}, \xi) = \widetilde{D}(\boldsymbol{\beta}, \boldsymbol{\gamma}, \xi)\widetilde{V}^{-1}\widetilde{S}(\boldsymbol{\beta}, \boldsymbol{\gamma}, \xi)$, where
\begin{align*}
\widetilde{D}(\boldsymbol{\beta}, \boldsymbol{\gamma}, \xi) &= \int_{\nu-(\gamma_{0}+\b... | 2507.09468v1_22.png |
\begin{figure}[t]
\centering
\includegraphics[width=\linewidth]{figure_5.pdf}
\caption{Overview of LGSP-Prompt framework. Our approach leverages both local and global spatial information through three key components: (1) Local Spatial Prompting with dynamic prompt selection for fine-grained feature extraction, (2) Glob... | 2507.09183v2_4.png |
\begin{table}[t]
\centering
\resizebox{\linewidth}{!}{
\begin{tabular}{lccccc}
\toprule
\multirow{2}{*}{\textbf{Model}} & \textbf{HQA} & \textbf{2Wiki} & \textbf{MATH} & \textbf{GSM8K} & \textbf{IFEval} \\
& \textbf{EM} & \textbf{EM} & \textbf{Acc} & \textbf{Acc} & \textbf{SAcc} \\
\midrule
\textbf{AutoTIR} & 43.15 & ... | 2507.21836v1_5.png |
differentiable.
\section{Problem Statement}
Optimal transport involves finding the most efficient way to transform one probability distribution into another. The distributions can be thought of as descriptions of how mass (representing probability) is spread over a space. Efficiency here is measured by a predefined co... | 2507.13191v2_1.png |
\begin{table}[t]
\centering
\caption{Comparison of different methods for cancer lesion segmentation with Weighted Averages. Bold values indicate the highest score, and underlined values indicate the second highest. All comparison results yielded $p<0.05$. ("Single" denotes single-task training, "+" indicates nnU-Net wi... | 2507.07126v1_6.png |
\section{References}
@article{1,
author = {Rosi{\'e} et al., V.},
title = {Perception and social evaluation of cloned and recorded voices: Effects of familiarity and self-relevance},
journal = {Computers in Human Behavior: Artificial Humans},
year = {2025}
}
@misc{2,
author = {Peek, S.},
title = {How digi... | 2510.23096v1_4.png |
overlooked, affecting comparison outcomes. Additionally, on average, the hybrid method takes 90 seconds to process a single report pair running on an NVIDIA GeForce RTX 2080 Ti, limiting its scalability for large-scale datasets.
Future work will focus on several directions: (1) Exploring how the choice of NER model af... | 2510.03102v2_11.png |
\begin{figure}[t]
\centering
\includegraphics[width=\linewidth]{figure_3.pdf}
\caption{Illustration of cluster Prototype-Guided cross-modal Alignment (PGA): (a) Instance-based image-text contrastive learning aims to draw positive pairs together while separating negative pairs. (b \& c) Cluster Prototype-Guided Cross-mo... | 2507.09256v1_7.png |
\title{AU-LLM: Micro-Expression Action Unit Detection via Enhanced LLM-Based Feature Fusion}
\author{Zhishu Liu$^{1\ast}$, Kaishen Yuan$^{1\ast}$, Bo Zhao$^{1}$, Yong Xu$^{2}$, and Zitong Yu$^{1(\boxtimes)}$\\
$^{1}$Great Bay University\\
$^{2}$Harbin Institute of Technology, Shenzhen}
\begin{abstract}
The detection ... | 2507.21778v1_0.png |
\begin{table*}[tbp]
\centering
\caption{Image-text retrieval performance comparison on Flickr30K and MSCOCO. $^\star$ and $^\dagger$ denote visual semantic description generated by Florence-2-large-ft-0.77B and MiniCPMV2.6-8B respectively, while * indicates ensemble model results.}
\label{tab:table_1}
\begin{tabular}{l... | 2507.08590v1_3.png |
\begin{figure}[t]
\centering
\includegraphics[width=\linewidth]{figure_1.pdf}
\caption{Design of PoliAnalyzer. Solid arrows represent data flow within PoliAnalyzer; dashed arrows represent main user flows.}
\label{figure_1}
\end{figure}
We use each \emph{line} of the privacy policy as a segment, as a balance between a... | 2507.14214v1_2.png |
\begin{table}[htbp]
\centering
\resizebox{\columnwidth}{!}{
\begin{tabular}{|l|r|r|r|r|l|}
\hline
\emph{dastgāh}/\emph{āvāz} & \emph{gūsheh} Count & Number of Notes & Total Duration (unit) & MIDI Performance Duration (second) & Pitch Range \\
\hline
\emph{Shur} & 29 & 4830 & 9839 & 1966 ... | 2507.10456v3_4.png |
\begin{table*}[htp]
\centering
\setlength{\tabcolsep}{3pt}
\begin{tabular}{c|cc|cc|cc|cc}
\toprule
\multirow{2}{*}{Approach} & \multicolumn{2}{c|}{Facescrub/ResNet} & \multicolumn{2}{c|}{SVHN/VGG} & \multicolumn{2}{c|}{CIFAR-10/ViT} & \multicolumn{2}{c}{CIFAR-100/ResNet}\\
& $Acc_{\mathcal{D}_{r... | 2507.21738v1_4.png |
\title{Revealing the Hidden Temporal Structure of HubertSoft Embeddings based on the Russian Phonetic Corpus}
\author{Anastasia Ananeva$^{1}$, Anton Tomilov$^{2}$, and Marina Volkova$^{2}$\\
$^{1}$ ITMO University, Saint Petersburg, Russia\\
$^{2}$ STC-innovations Ltd., Russia\\
{\tt ananeva@itmo.ru}\\
{\tt tomilov, v... | 2507.06794v1_0.png |
\begin{table}[t]
\centering
\caption{Top-1 Accuracy Comparison of Token Compression Methods on AutoFormer-S in Off-the-Shelf and Retrained Settings (Acc1(ots) vs. Acc1(re-train)), with Inference Efficiency Measured by GFLOPs and Throughput (img/s).}
\label{table_2}
\begin{tabular}{lcccc}
\hline
Method & Acc1(ots)$\upar... | 2507.09702v1_3.png |
relation-aware attention mechanism \cite{28}, HRAN aggregates features along relational paths \cite{21}, and KGT5 and HittER leverage Transformers for scalable and context-aware representation learning \cite{27,4}. Despite their success, embedding-based models still struggle with distinguishing semantically similar but... | 2507.20643v2_4.png |
(BTA) matrix. While $\Q_c$ stores information on the conditional dependencies, its inverse $\boldsymbol{\Sigma}_c = \Q_c^{-1}$ contains information on the linear dependence between two variables in the off-diagonal entries, while its diagonal entries store the variances of the individual variables which are of interest... | 2507.06938v2_3.png |
Starobinsky gravity, both neutron stars~\cite{37} and black holes~\cite{42} can support massive scalar hair, with their exterior spacetime exhibiting nonzero curvature, while the Schwarzschild black hole remains a special case with vanishing exterior curvature. (Similar types of hairy solutions also arise in general qu... | 2507.18916v1_2.png |
\begin{figure}[t]
\centering
\begin{subfigure}{0.45\linewidth}
\centering
\includegraphics[width=\linewidth]{figure_4_1.pdf}
\caption{Valencian Region}
\end{subfigure}
\hfill
\begin{subfigure}{0.45\linewidth}
\centering
\includegraphics[width=\linewidth]{figure_4_2.pdf}
\caption{Castile and Leon}
\end{subfigure}
\vspa... | 2507.08376v1_13.png |
the concept of grain size, which defines the level at which engagement is conceptualized and measured. Grain size ranges from macro-level (e.g., group engagement) to micro-level (e.g., momentary individual engagement in a specific task). Micro-level engagement can be assessed using physiological signals such as blink r... | 2507.17959v1_1.png |
\title{HH-Codec: High Compression High-fidelity Discrete Neural Codec for Spoken Language Modeling}
\author{Rongkun Xue$^{1}$ \and Yazhe Niu$^{2,3}$ \and Shuai Hu$^{4}$ \and Zixin Yin$^{4}$ \and Yongqiang Yao$^{5}$ \and Jing Yang$^{1}$}
\begin{abstract}
Discrete speech tokenization is a fundamental component in speec... | 2507.18897v1_0.png |
\begin{figure}[t]
\centering
\includegraphics[width=\linewidth]{figure_3.pdf}
\caption{The diurnal variation of AfriTEC model with F10.7 and Dst index value comparision for model evaluation with GNSS data in MAL2 station in October 12--20, 2016.}
\label{figure_3}
\end{figure}
\begin{figure}[t]
\centering
\includegraph... | 2507.10275v1_9.png |
End of preview. Expand in Data Studio
TexOCR: Benchmarking and Advancing Document OCR Models for Compilable Page-to-LaTeX Reconstruction [ACL 2026 Main]
This repository provides the RL training data in JSON format for TexOCR.
Overview
- Dataset: TexOCR_RL_json
- Type: Training Data (JSON)
- Training Method: GRPO (Reinforcement Learning)
- Task: Document OCR → LaTeX generation
- Downloads last month
- 29