To build a math MLLM, current research efforts predominantly rely on elaborate training schemes or high-quality multi-modal math reasoning data. On the other hand, model merging has attracted much attention recently due to its low-cost and tune-free property, which has shown great potential for constructing math MLLM. However, most recent attempts either assume all parameters are task-relevant or adopt a less interpretable parameter selection strategy, thus merging parameters on a coarse granularity, leading to limited or even degrading performance. In this paper, we propose a causal-analysis–guided parameter merging framework, which identifies the math reasoning-related parameters in a finer granularity and interpretable manner, then merges the math LLMs with MLLM. Specifically, we first construct targeted arithmetic and logical tasks, generate reference and counterfactual samples, and then apply intervention-based causal attribution to localize parameters that are causally responsible for math reasoning. Once we identify the crucial reasoning parameters in the math LLM that exhibit higher causal effects, we can more precisely align and merge only the corresponding parameters from the math LLM into the base MLLM. The procedure is tuning-free and enhances both math reasoning and general reasoning abilities. Experimental results of three model families on four math reasoning benchmarks and two general reasoning benchmarks validate the effectiveness of the proposed approach.

Plane geometry problem (PGP) solving has become a critical benchmark for multimodal reasoning because it requires accurate visual perception and precise multi-step symbolic deduction. Although test-time scaling (TTS) has demonstrated remarkable success in general mathematical reasoning, it fails to scale effectively under the symbolic-program paradigm for plane geometry. We identify two key obstacles: limited reasoning diversity induced by rigid symbolic programs and insufficient explicit visual grounding before symbolic deduction. To address these issues, we propose Multi-Trace Synthesis (MTS), which converts each symbolic program into heterogeneous reasoning traces, including executable Python scripts and CoT-augmented variants. We further propose Perception-Augmented (PA) training, which parses diagrams into structured semantic clauses before deduction, and Consensus-Guided Multi-Trace Ensemble (CG-MTE) for efficient self-adaptive inference. Experiments on three geometry benchmarks show that our method consistently improves PGP-solving across model scales and achieves strong performance against both general-purpose MLLMs and specialized geometry solvers. Under test-time scaling, CG-MTE achieves a comparable accuracy to high-budget self-consistency while reducing sampling cost by up to 8×. Code and data are publicly available at https://github.com/Jason8Kang/ReTTS-PGPS.

As large language models (LLMs) transition to autonomous agents synthesizing real-time information, their reasoning capabilities introduce an unexpected attack surface. This paper introduces a novel threat where colluding agents steer victim beliefs using only truthful evidence fragments distributed through public channels, without relying on covert communications, backdoors, or falsified documents. By exploiting LLMs’ overthinking tendency, we formalize the first cognitive collusion attack and propose Generative Montage: a Writer-Editor-Director framework that constructs deceptive narratives through adversarial debate and coordinated posting of evidence fragments, causing victims to internalize and propagate fabricated conclusions. To study this risk, we develop CoPHEME, a dataset derived from real-world rumor events, and simulate attacks across diverse LLM families. Our results show pervasive vulnerability across 14 LLM families: attack success rates reach 74.4% for proprietary models and 70.6% for open-weights models. Counterintuitively, stronger reasoning capabilities increase susceptibility, with reasoning-specialized models showing higher attack success than base models or prompts. Furthermore, these false beliefs then cascade to downstream judges, achieving over 60% deception rates, highlighting a socio-technical vulnerability in how LLM-based agents interact with dynamic information environments.

This paper primarily demonstrates a method to quantitatively assess the alignment between multi-step, structured reasoning in large language models and human preferences. We introduce the Alignment Score, a semantic-level metric that compares a model-produced chain of thought traces with a human-preferred reference by constructing semantic-entropy-based matrices over intermediate steps and measuring their divergence. Our analysis shows that Alignment Score tracks task accuracy across models and hop depths, and peaks at 2-hop reasoning. Empirical results further indicate that misalignment at greater reasoning depths is driven mainly by alignment errors such as thematic shift and redundant reasoning. Viewing chain sampling as drawing from a distribution over reasoning paths, we empirically demonstrate a strong and consistent correlation between Alignment Score and accuracy, readability, and coherence, supporting its use as a diagnostic signal. The code is available.

Existing table understanding methods face challenges due to complex table structures and intricate logical reasoning. While supervised finetuning (SFT) dominates existing research, reinforcement learning (RL), such as Group Relative Policy Optimization (GRPO), has shown promise but struggled with low initial policy accuracy and coarse rewards in tabular contexts. In this paper, we introduce Table-R1, a three-stage RL framework that enhances multimodal table understanding through: (1) Warm-up that prompts initial perception and reasoning capabilities, (2) Perception Alignment GRPO (PA-GRPO), which employs continuous Tree-Edit-Distance Similarity (TEDS) rewards for recognizing table structures and contents, and (3) Hint-Completion GRPO (HC-GRPO), which utilizes fine-grained rewards of residual steps based on the hint-guided question. Extensive experiments demonstrate that Table-R1 can boost the model’s table reasoning performance obviously on both held-in and held-out datasets, outperforming SFT and GRPO largely. Notably, Qwen2-VL-7B with Table-R1 surpasses larger specific table understanding models (e.g., Table-LLaVA 13B), even achieving comparable performance to the closed-source model GPT-4o on held-in datasets, demonstrating the efficacy of each stage of Table-R1 in overcoming initialization bottlenecks and reward sparsity, thereby advancing robust multimodal table understanding.

Solving math word problem (MWP) with AI techniques has recently made great progress with the success of deep neural networks (DNN), but it is far from being solved. We argue that the ability of learning by analogy is essential for an MWP solver to better understand same problems which may typically be formulated in diverse ways. However most existing works exploit the shortcut learning to train MWP solvers simply based on samples with a single question. In lack of diverse questions, these methods merely learn shallow heuristics. In this paper, we make a first attempt to solve MWPs by generating diverse yet consistent questions/equations. Given a typical MWP including the scenario description, question, and equation (i.e., answer), we first generate multiple consistent equations via a group of heuristic rules. We then feed them to a question generator together with the scenario to obtain the corresponding diverse questions, forming a new MWP with a variety of questions and equations. Finally we engage a data filter to remove those unreasonable MWPs, keeping the high-quality augmented ones. To evaluate the ability of learning by analogy for an MWP solver, we generate a new MWP dataset (called DiverseMath23K) with diverse questions by extending the current benchmark Math23K. Extensive experimental results demonstrate that our proposed method can generate high-quality diverse questions with corresponding equations, further leading to performance improvement on Diverse-Math23K. The code and dataset is available at: https://github.com/zhouzihao501/DiverseMWP.

This is the system description of the Harbin Institute of Technology Shenzhen (HITSZ) team for the first and second subtasks of the fourth Social Media Mining for Health Applications (SMM4H) shared task in 2019. The two subtasks are automatic classification and extraction of adverse effect mentions in tweets. The systems for the two subtasks are based on bidirectional encoder representations from transformers (BERT), and achieves promising results. Among the systems we developed for subtask1, the best F1-score was 0.6457, for subtask2, the best relaxed F1-score and the best strict F1-score were 0.614 and 0.407 respectively. Our system ranks first among all systems on subtask1.