The advent of large reasoning models, such as OpenAI o1 and DeepSeek R1, has significantly advanced complex reasoning tasks. However, their capabilities in multilingual complex reasoning remain underexplored, with existing efforts largely focused on simpler tasks like MGSM. To address this gap, we introduce MMATH, a benchmark for multilingual complex reasoning spanning 374 high-quality math problems across 10 typologically diverse languages. Using MMATH, we observe that even advanced models like DeepSeek R1 exhibit substantial performance disparities across languages and suffer from a critical off-target issue-generating responses in unintended languages. To address this, we explore strategies including prompting and training, demonstrating that reasoning in English and answering in target languages can simultaneously enhance performance and preserve target-language consistency. Our findings offer new insights and practical strategies for advancing the multilingual reasoning capabilities of large language models. Our code and data could be found at https://github.com/RUCAIBox/MMATH.

Recent advancements in Large Vision-Language Models (VLMs), have greatly enhanced their capability to jointly process text and images. However, despite extensive benchmarks evaluating visual comprehension (e.g., diagrams, color schemes, OCR tasks...), there is limited assessment of VLMs' ability to read and reason about text-rich images effectively. To fill this gap, we introduce ReadBench, a multimodal benchmark specifically designed to evaluate the reading comprehension capabilities of VLMs. ReadBench transposes contexts from established text-only benchmarks into images of text while keeping textual prompts and questions intact. Evaluating leading VLMs with ReadBench, we find minimal-but-present performance degradation on short, text-image inputs, while performance sharply declines for longer, multi-page contexts. Our experiments further reveal that text resolution has negligible effects on multimodal performance. These findings highlight needed improvements in VLMs, particularly their reasoning over visually presented extensive textual content, a capability critical for practical applications. ReadBench is available at https://github.com/answerdotai/ReadBench .

Applying Reinforcement Learning (RL) to Video Large Language Models (Video-LLMs) shows significant promise for complex video reasoning. However, popular Reinforcement Fine-Tuning (RFT) methods, such as outcome-based Group Relative Policy Optimization (GRPO), are limited by data preparation bottlenecks (e.g., noise or high cost) and exhibit unstable improvements in the quality of long chain-of-thoughts (CoTs) and downstream performance.To address these limitations, we propose VerIPO, a Verifier-guided Iterative Policy Optimization method designed to gradually improve video LLMs' capacity for generating deep, long-term reasoning chains. The core component is Rollout-Aware Verifier, positioned between the GRPO and Direct Preference Optimization (DPO) training phases to form the GRPO-Verifier-DPO training loop. This verifier leverages small LLMs as a judge to assess the reasoning logic of rollouts, enabling the construction of high-quality contrastive data, including reflective and contextually consistent CoTs. These curated preference samples drive the efficient DPO stage (7x faster than GRPO), leading to marked improvements in reasoning chain quality, especially in terms of length and contextual consistency. This training loop benefits from GRPO's expansive search and DPO's targeted optimization. Experimental results demonstrate: 1) Significantly faster and more effective optimization compared to standard GRPO variants, yielding superior performance; 2) Our trained models exceed the direct inference of large-scale instruction-tuned Video-LLMs, producing long and contextually consistent CoTs on diverse video reasoning tasks; and 3) Our model with one iteration outperforms powerful LMMs (e.g., Kimi-VL) and long reasoning models (e.g., Video-R1), highlighting its effectiveness and stability.

We introduce a novel design of parallel-jaw grippers drawing inspiration from pin-pression toys. The proposed pin-pression gripper features a distinctive mechanism in which each finger integrates a 2D array of pins capable of independent extension and retraction. This unique design allows the gripper to instantaneously customize its finger's shape to conform to the object being grasped by dynamically adjusting the extension/retraction of the pins. In addition, the gripper excels in in-hand re-orientation of objects for enhanced grasping stability again via dynamically adjusting the pins. To learn the dynamic grasping skills of pin-pression grippers, we devise a dedicated reinforcement learning algorithm with careful designs of state representation and reward shaping. To achieve a more efficient grasp-while-lift grasping mode, we propose a curriculum learning scheme. Extensive evaluations demonstrate that our design, together with the learned skills, leads to highly flexible and robust grasping with much stronger generality to unseen objects than alternatives. We also highlight encouraging physical results of sim-to-real transfer on a physically manufactured pin-pression gripper, demonstrating the practical significance of our novel gripper design and grasping skill. Demonstration videos for this paper are available at https://github.com/siggraph-pin-pression-gripper/pin-pression-gripper-video.

--With the advancement of edge computing, federated learning (FL) displays a bright promise as a privacy-preserving collaborative learning paradigm. However, one major challenge for FL is the data heterogeneity issue, which refers to the biased labeling preferences among multiple clients, negatively impacting convergence and model performance. Most previous FL methods attempt to tackle the data heterogeneity issue locally or globally, neglecting underlying class-wise structure information contained in each client. In this paper, we first study how data heterogeneity affects the divergence of the model and decompose it into local, global, and sampling drift sub-problems. T o explore the potential of using intra-client class-wise structural knowledge in handling these drifts, we thus propose Federated Learning with Structural Knowledge Collaboration (FedSKC). The key idea of FedSKC is to extract and transfer domain preferences from inter-client data distributions, offering diverse class-relevant knowledge and a fair convergent signal. FedSKC comprises three components: i) local contrastive learning, to prevent weight divergence resulting from local training; ii) global discrepancy aggregation, which addresses the parameter deviation between the server and clients; iii) global period review, correcting for the sampling drift introduced by the server randomly selecting devices. We have theoretically analyzed FedSKC under non-convex objectives and empirically validated its superiority through extensive experimental results.

This work studies knowledge distillation (KD) for large language models (LLMs) through preference optimization. We propose a reward-guided imitation learning framework for sequential KD, formulating a min-max optimization problem between the policy and reward model (RM) to minimize the performance gap between the student and teacher policies. Specifically, the reward optimization is constrained to achieve near-optimality within a confidence set for preference alignment. For preference data construction, we explore both offline and online preference-based KD. Additionally, we reformulate the RM using the $Q$-value function and extend the framework to white-box KD, where the teacher policy's predicted probabilities are accessible. Theoretical analysis and empirical results demonstrate the effectiveness of the proposed framework.

We introduce the Exemplar-Based Expository Text Generation task, aiming to generate an expository text on a new topic using an exemplar on a similar topic. Current methods fall short due to their reliance on extensive exemplar data, difficulty in adapting topic-specific content, and issues with long-text coherence. To address these challenges, we propose the concept of Adaptive Imitation and present a novel Recurrent Plan-then-Adapt (RePA) framework. RePA leverages large language models (LLMs) for effective adaptive imitation through a fine-grained plan-then-adapt process. RePA also enables recurrent segment-by-segment imitation, supported by two memory structures that enhance input clarity and output coherence. We also develop task-specific evaluation metrics--imitativeness, adaptiveness, and adaptive-imitativeness--using LLMs as evaluators. Experimental results across our collected three diverse datasets demonstrate that RePA surpasses existing baselines in producing factual, consistent, and relevant texts for this task.

Background knowledge is typically needed for successful comprehension of topical and domain specific reading passages, such as in the STEM domain. However, there are few automated measures of student knowledge that can be readily deployed and scored in time to make predictions on whether a given student will likely be able to understand a specific content area text. In this paper, we present our effort in developing K-tool, an automated system for generating topical vocabulary tests that measure students' background knowledge related to a specific text. The system automatically detects the topic of a given text and produces topical vocabulary items based on their relationship with the topic. This information is used to automatically generate background knowledge forms that contain words that are highly related to the topic and words that share similar features but do not share high associations to the topic. Prior research indicates that performance on such tasks can help determine whether a student is likely to understand a particular text based on their knowledge state. The described system is intended for use with middle and high school student population of native speakers of English. It is designed to handle single reading passages and is not dependent on any corpus or text collection. In this paper, we describe the system architecture and present an initial evaluation of the system outputs.

Data-centric distillation, including data augmentation, selection, and mixing, offers a promising path to creating smaller, more efficient student Large Language Models (LLMs) that retain strong reasoning abilities. However, there still lacks a comprehensive benchmark to systematically assess the effect of each distillation approach. This paper introduces DC-CoT, the first data-centric benchmark that investigates data manipulation in chain-of-thought (CoT) distillation from method, model and data perspectives. Utilizing various teacher models (e.g., o4-mini, Gemini-Pro, Claude-3.5) and student architectures (e.g., 3B, 7B parameters), we rigorously evaluate the impact of these data manipulations on student model performance across multiple reasoning datasets, with a focus on in-distribution (IID) and out-of-distribution (OOD) generalization, and cross-domain transfer. Our findings aim to provide actionable insights and establish best practices for optimizing CoT distillation through data-centric techniques, ultimately facilitating the development of more accessible and capable reasoning models. The dataset can be found at https://huggingface.co/datasets/rana-shahroz/DC-COT, while our code is shared in https://anonymous.4open.science/r/DC-COT-FF4C/.

Direct Preference Optimization (DPO) has emerged as a promising framework for aligning Large Language Models (LLMs) with human preferences by directly optimizing the log-likelihood difference between chosen and rejected responses. However, existing methods assign equal importance to all tokens in the response, while humans focus on more meaningful parts. This leads to suboptimal preference optimization, as irrelevant or noisy tokens disproportionately influence DPO loss. To address this limitation, we propose \textbf{O}ptimal \textbf{T}ransport-based token weighting scheme for enhancing direct \textbf{P}reference \textbf{O}ptimization (OTPO). By emphasizing semantically meaningful token pairs and de-emphasizing less relevant ones, our method introduces a context-aware token weighting scheme that yields a more contrastive reward difference estimate. This adaptive weighting enhances reward stability, improves interpretability, and ensures that preference optimization focuses on meaningful differences between responses. Extensive experiments have validated OTPO's effectiveness in improving instruction-following ability across various settings\footnote{Code is available at https://github.com/Mimasss2/OTPO.}.