Vision Matters: Simple Visual Perturbations Can Boost Multimodal Math Reasoning

Despite the rapid progress of multimodal large language models (MLLMs), they have largely overlooked the importance of visual processing. In a simple yet revealing experiment, we interestingly find that language-only models, when provided with image captions, can achieve comparable or even better performance than MLLMs that consume raw visual inputs. This suggests that current MLLMs may generate accurate visual descriptions but fail to effectively integrate them during reasoning. Motivated by this, we propose a simple visual perturbation framework that enhances perceptual robustness without requiring algorithmic modifications or additional training data. Our approach introduces three targeted perturbations: distractor concatenation, dominance-preserving mixup, and random rotation, that can be easily integrated into existing post-training pipelines including SFT, DPO, and GRPO. Through extensive experiments across multiple datasets, we demonstrate consistent improvements in mathematical reasoning performance, with gains comparable to those achieved through algorithmic changes. Additionally, we achieve competitive performance among open-source 7B RL-tuned models by training Qwen2.5-VL-7B with visual perturbation. Through comprehensive ablation studies, we analyze the effectiveness of different perturbation strategies, revealing that each perturbation type contributes uniquely to different aspects of visual reasoning. Our findings highlight the critical role of visual perturbation in multimodal mathematical reasoning: better reasoning begins with better seeing. Our code is available at https://github.com/YutingLi0606/Vision-Matters.

Advancing Multimodal Reasoning via Reinforcement Learning with Cold Start

Recent advancements in large language models (LLMs) have demonstrated impressive chain-of-thought reasoning capabilities, with reinforcement learning (RL) playing a crucial role in this progress. While "aha moment" patterns--where models exhibit self-correction through reflection--are often attributed to emergent properties from RL, we first demonstrate that these patterns exist in multimodal LLMs (MLLMs) prior to RL training but may not necessarily correlate with improved reasoning performance. Building on these insights, we present a comprehensive study on enhancing multimodal reasoning through a two-stage approach: (1) supervised fine-tuning (SFT) as a cold start with structured chain-of-thought reasoning patterns, followed by (2) reinforcement learning via GRPO to further refine these capabilities. Our extensive experiments show that this combined approach consistently outperforms both SFT-only and RL-only methods across challenging multimodal reasoning benchmarks. The resulting models achieve state-of-the-art performance among open-source MLLMs at both 3B and 7B scales, with our 7B model showing substantial improvements over base models (e.g., 66.3 %$\rightarrow$73.4 % on MathVista, 62.9 %$\rightarrow$70.4 % on We-Math) and our 3B model achieving performance competitive with several 7B models. Overall, this work provides practical guidance for building advanced multimodal reasoning models. Our code is available at https://github.com/waltonfuture/RL-with-Cold-Start.

Unsupervised Post-Training for Multi-Modal LLM Reasoning via GRPO

Improving Multi-modal Large Language Models (MLLMs) in the post-training stage typically relies on supervised fine-tuning (SFT) or reinforcement learning (RL). However, these supervised methods require expensive and manually annotated multi-modal data--an ultimately unsustainable resource. While recent efforts have explored unsupervised post-training, their methods are complex and difficult to iterate. In this work, we are the first to investigate the use of GRPO, a stable and scalable online RL algorithm, for enabling continual self-improvement without any external supervision. We propose MM-UPT, a simple yet effective framework for unsupervised post-training of MLLMs. MM-UPT builds upon GRPO, replacing traditional reward signals with a self-rewarding mechanism based on majority voting over multiple sampled responses. Our experiments demonstrate that MM-UPT significantly improves the reasoning ability of Qwen2.5-VL-7B (e.g., 66.3 %$\rightarrow$72.9 % on MathVista, 62.9 %$\rightarrow$68.7 % on We-Math), using standard dataset without ground truth labels. MM-UPT also outperforms prior unsupervised baselines and even approaches the results of supervised GRPO. Furthermore, we show that incorporating synthetic questions, generated solely by MLLM itself, can boost performance as well, highlighting a promising approach for scalable self-improvement. Overall, MM-UPT offers a new paradigm for continual, autonomous enhancement of MLLMs in the absence of external supervision. Our code is available at https://github.com/waltonfuture/MM-UPT.

ConCISE: Confidence-guided Compression in Step-by-step Efficient Reasoning

Large Reasoning Models (LRMs) perform strongly in complex reasoning tasks via Chain-of-Thought (CoT) prompting, but often suffer from verbose outputs caused by redundant content, increasing computational overhead, and degrading user experience. Existing compression methods either operate post-hoc pruning, risking disruption to reasoning coherence, or rely on sampling-based selection, which fails to intervene effectively during generation. In this work, we introduce a confidence-guided perspective to explain the emergence of redundant reflection in LRMs, identifying two key patterns: Confidence Deficit, where the model reconsiders correct steps due to low internal confidence, and Termination Delay, where reasoning continues even after reaching a confident answer. Based on this analysis, we propose ConCISE (Confidence-guided Compression In Step-by-step Efficient Reasoning), a framework that simplifies reasoning chains by reinforcing the model's confidence during inference, thus preventing the generation of redundant reflection steps. It integrates Confidence Injection to stabilize intermediate steps and Early Stopping to terminate reasoning when confidence is sufficient. Extensive experiments demonstrate that fine-tuning LRMs on ConCISE-generated data yields significantly shorter outputs, reducing length by up to approximately 50% under SimPO, while maintaining high task accuracy. ConCISE consistently outperforms existing baselines across multiple reasoning benchmarks.

Ensuring Force Safety in Vision-Guided Robotic Manipulation via Implicit Tactile Calibration

In dynamic environments, robots often encounter constrained movement trajectories when manipulating objects with specific properties, such as doors. Therefore, applying the appropriate force is crucial to prevent damage to both the robots and the objects. However, current vision-guided robot state generation methods often falter in this regard, as they lack the integration of tactile perception. To tackle this issue, this paper introduces a novel state diffusion framework termed SafeDiff. It generates a prospective state sequence from the current robot state and visual context observation while incorporating real-time tactile feedback to refine the sequence. As far as we know, this is the first study specifically focused on ensuring force safety in robotic manipulation. It significantly enhances the rationality of state planning, and the safe action trajectory is derived from inverse dynamics based on this refined planning. In practice, unlike previous approaches that concatenate visual and tactile data to generate future robot state sequences, our method employs tactile data as a calibration signal to adjust the robot's state within the state space implicitly. Additionally, we've developed a large-scale simulation dataset called SafeDoorManip50k, offering extensive multimodal data to train and evaluate the proposed method. Extensive experiments show that our visual-tactile model substantially mitigates the risk of harmful forces in the door opening, across both simulated and real-world settings.

Contextual Bandits with Arm Request Costs and Delays

We introduce a novel extension of the contextual bandit problem, where new sets of arms can be requested with stochastic time delays and associated costs. In this setting, the learner can select multiple arms from a decision set, with each selection taking one unit of time. The problem is framed as a special case of semi-Markov decision processes (SMDPs). The arm contexts, request times, and costs are assumed to follow an unknown distribution. We consider the regret of an online learning algorithm with respect to the optimal policy that achieves the maximum average reward. By leveraging the Bellman optimality equation, we design algorithms that can effectively select arms and determine the appropriate time to request new arms, thereby minimizing their regret. Under the realizability assumption, we analyze the proposed algorithms and demonstrate that their regret upper bounds align with established results in the contextual bandit literature. We validate the algorithms through experiments on simulated data and a movie recommendation dataset, showing that their performance is consistent with theoretical analyses.

MC-CoT: A Modular Collaborative CoT Framework for Zero-shot Medical-VQA with LLM and MLLM Integration

In recent advancements, multimodal large language models (MLLMs) have been fine-tuned on specific medical image datasets to address medical visual question answering (Med-VQA) tasks. However, this common approach of task-specific fine-tuning is costly and necessitates separate models for each downstream task, limiting the exploration of zero-shot capabilities. In this paper, we introduce MC-CoT, a modular cross-modal collaboration Chain-of-Thought (CoT) framework designed to enhance the zero-shot performance of MLLMs in Med-VQA by leveraging large language models (LLMs). MC-CoT improves reasoning and information extraction by integrating medical knowledge and task-specific guidance, where LLM provides various complex medical reasoning chains and MLLM provides various observations of medical images based on instructions of the LLM. Our experiments on datasets such as SLAKE, VQA-RAD, and PATH-VQA show that MC-CoT surpasses standalone MLLMs and various multimodality CoT frameworks in recall rate and accuracy. These findings highlight the importance of incorporating background information and detailed guidance in addressing complex zero-shot Med-VQA tasks.

Prototype of Secure Wire-Line Telephone

This paper proposes a secure wire-line telephone prototype that leverages physical layer security (PLS) techniques to protect communications from wiretapping. The system generates artificial noise (AN) in both directions over a telephone line and utilizes a telephone hybrid circuit to achieve effective AN cancellation. We conduct a thorough analysis of the secrecy capacity and evaluate the system's performance through both simulations and practical experiments. The results demonstrate that the proposed scheme significantly enhances communication security while preserving the integrity of legitimate signals, making it a robust and viable solution for secure telephone systems.

Can I understand what I create? Self-Knowledge Evaluation of Large Language Models

Large language models (LLMs) have achieved remarkable progress in linguistic tasks, necessitating robust evaluation frameworks to understand their capabilities and limitations. Inspired by Feynman's principle of understanding through creation, we introduce a self-knowledge evaluation framework that is easy to implement, evaluating models on their ability to comprehend and respond to self-generated questions. Our findings, based on testing multiple models across diverse tasks, reveal significant gaps in the model's self-knowledge ability. Further analysis indicates these gaps may be due to misalignment with human attention mechanisms. Additionally, fine-tuning on self-generated math task may enhance the model's math performance, highlighting the potential of the framework for efficient and insightful model evaluation and may also contribute to the improvement of LLMs.

Adaptive Online Experimental Design for Causal Discovery

Causal discovery aims to uncover cause-and-effect relationships encoded in causal graphs by leveraging observational, interventional data, or their combination. The majority of existing causal discovery methods are developed assuming infinite interventional data. We focus on data interventional efficiency and formalize causal discovery from the perspective of online learning, inspired by pure exploration in bandit problems. A graph separating system, consisting of interventions that cut every edge of the graph at least once, is sufficient for learning causal graphs when infinite interventional data is available, even in the worst case. We propose a track-and-stop causal discovery algorithm that adaptively selects interventions from the graph separating system via allocation matching and learns the causal graph based on sampling history. Given any desired confidence value, the algorithm determines a termination condition and runs until it is met. We analyze the algorithm to establish a problem-dependent upper bound on the expected number of required interventional samples. Our proposed algorithm outperforms existing methods in simulations across various randomly generated causal graphs. It achieves higher accuracy, measured by the structural hamming distance (SHD) between the learned causal graph and the ground truth, with significantly fewer samples.