Ziyu Chen, Yilun Zhao, Chengye Wang, Rilyn R. Han, Manasi Patwardhan, Arman Cohan. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Existing document OCR largely targets plain text or Markdown, discarding the structural and executable properties that make LaTeX essential for scientific publishing.We study page-level reconstruction of scientific PDFs into compilable LaTeX and introduce TEX-OCR-Bench, a benchmark, and TEXOCR-Train, a large-scale training corpus, for this task.TEXOCR-Bench features a multi-dimensional evaluation suite that jointly assesses transcription fidelity, structural faithfulness, and endto-end compilability.Leveraging TEXOCR-Train, we train a 2B-parameter model, TEX-OCR, using supervised fine-tuning (SFT) and reinforcement learning (RL) with verifiable rewards derived from LaTeX unit tests that directly enforce compilability and referential integrity.Experiments across 21 frontier models on TEXOCR-Bench show that existing systems frequently violate key document invariants, including consistent section structure, correct float placement, and valid label-reference links, which undermines compilation reliability and downstream usability.Our analysis further reveals that RL with verifiable rewards yields consistent improvements over SFT alone, particularly on structural and compilation metrics.

Trajectory planning is essential for ensuring the safe operation of autonomous vehicles. However, existing methods rarely consider the vehicle’s multi-coupled dynamics, including lateral-longitudinal motion coupling, tire force coupling, and lateral instability. This omission can result in infeasible trajectories, vehicle instability, or even accidents under extreme conditions. To address this challenge, this study presents a multi-coupled dynamics trajectory planning (MCTP) scheme. MCTP establishes a coupled kinematics model to accurately represent vehicle motion states and constructs a tire force representation model, which based solely on vehicle motion states, facilitating seamless integration into trajectory planning. By incorporating coupled tire force characteristics and lateral stability analysis, a set of coupled dynamic constraints is formulated to ensure trajectory feasibility and lateral stability. Additionally, a multi-objective function is designed to further optimize trajectory safety, dynamic feasibility, and lateral stability, with the optimal trajectory obtained through receding horizon optimization. Closed-loop validation on both hardware-in-the-loop and real-vehicle experimental platforms demonstrates that, MCTP generates trajectories with enhanced safety and feasibility. It also improves tracking stability margins and dynamics performance, highlighting its effectiveness in handling extreme conditions.

Yilun Zhao, Weiyuan Chen, Zhijian Xu, Manasi Patwardhan, Chengye Wang, Yixin Liu, Lovekesh Vig, Arman Cohan. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

We introduce SCIVER, the first benchmark specifically designed to evaluate the ability of foundation models to verify claims within a multimodal scientific context.SCIVER consists of 3,000 expert-annotated examples over 1,113 scientific papers, covering four subsets, each representing a common reasoning type in multimodal scientific claim verification.To enable fine-grained evaluation, each example includes expert-annotated supporting evidence.We assess the performance of 21 state-of-the-art multimodal foundation models, including o4mini, Gemini-2.5-Flash,Llama-3.2-Vision, and Qwen2.5-VL.Our experiment reveals a substantial performance gap between these models and human experts on SCIVER.Through an in-depth analysis of retrieval-augmented generation (RAG), and human-conducted error evaluations, we identify critical limitations in current open-source models, offering key insights to advance models' comprehension and reasoning in multimodal scientific literature tasks.

We introduce $\color{Blue}{\text{MMVU}}$, a comprehensive expert-level, multi-discipline benchmark for evaluating foundation models in video understanding. $\color{Blue}{\text{MMVU}}$ includes 3,000 expert-annotated questions spanning 27 subjects across four core disciplines: Science, Healthcare, Humanities & Social Sciences, and Engineering. Compared to prior benchmarks, $\color{Blue}{\text{MMVU}}$ features three key advancements. First, it challenges models to apply domain-specific knowledge and perform expert-level reasoning to analyze specialized-domain videos, moving beyond the basic visual perception typically assessed in current video benchmarks. Second, each example is annotated by human experts from scratch. We implement strict data quality controls to ensure the high quality of the dataset. Finally, each example is enriched with expert-annotated reasoning rationals and relevant domain knowledge, facilitating in-depth analysis. We conduct an extensive evaluation of 36 frontier multimodal foundation models on $\color{Blue}{\text{MMVU}}$. The latest System-2-capable models, o1 and Gemini 2.0 Flash Thinking, achieve the highest performance among the tested models. However, they still fall short of matching human expertise. Through in-depth error analyses and case studies, we offer actionable insights for future advancements in expert-level, knowledge-intensive video understanding for specialized domains.

Although Multimodal Large Language Models (MLLMs) have advanced numerous fields, their training on extensive multimodal datasets introduces significant privacy concerns, prompting the necessity for efficient unlearning methods.However, current multimodal unlearning approaches often directly adapt techniques from unimodal contexts, largely overlooking the critical issue of modality alignment, i.e., consistently removing knowledge across both unimodal and multimodal settings. To close this gap, we introduce UMU-bench, a unified benchmark specifically targeting modality misalignment in multimodal unlearning. UMU-bench consists of a meticulously curated dataset featuring 653 individual profiles, each described with both unimodal and multimodal knowledge.Additionally, novel tasks and evaluation metrics focusing on modality alignment are introduced, facilitating a comprehensive analysis of unimodal and multimodal unlearning effectiveness. Through extensive experimentation with state-of-the-art unlearning algorithms on UMU-bench, we demonstrate prevalent modality misalignment issues in existing methods. These findings underscore the critical need for novel multimodal unlearning approaches explicitly considering modality alignment.

Terrestrial laser scanning (TLS) has revolutionized tree-level measurement, enabling accurate trunk, and branch analysis, but it faces challenges in counting and measuring individual leaves in broad-leaved trees. We introduce TLSLeaf, an innovative method designed to accurately measure leaf count and area, specifically tailored for complex tree canopies. TLSLeaf addresses critical gaps in TLS-based leaf analysis through a four-step process: 1) wood-leaf separation; 2) individual leaf segmentation; 3) detection and repair of incomplete leaves; and 4) leaf counting and area measurement. TLSLeaf integrates graph theory and shortest-path algorithms for effective branch-leaf separation, followed by instance segmentation using a similarity graph. To overcome the challenge of TLS canopy occlusion and imperfect leaf scans, incomplete leaves are detected and repaired based on symmetry and concavity principles. TLSLeaf’s accuracy was validated through field scanning, in situ measurements, and destructive sampling. TLSLeaf showed a percentage error of 1.49%–8.60% for leaf counts (ranging from 201 to 4000 leaves) and achieved an$R^{2}$of 0.95 and RMSE of 5.87 cm2 for leaf areas (ranging from 9.98 to 179.77 cm2). TLSLeaf integrates semantic and instance segmentation through an unsupervised framework, presenting a “white-box” approach that ensures transparency and reproducibility. TLSLeaf represents a significant advancement in leaf-level analysis of TLS point clouds, with potential applications in enhancing tree reconstruction, 3-D radiative transfer modeling, and canopy photosynthesis.

Yilun Zhao, Yitao Long, Tintin Jiang, Chengye Wang, Weiyuan Chen, Hongjun Liu, Xiangru Tang, Yiming Zhang, Chen Zhao, Arman Cohan. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Multispectral images can be effectively smoothed by using the global distribution of pixel values to guide a local selective averaging process. After several iterations of this process, a typical image is virtually segmented into regions of constant value while significant edges in the image are preserved.