Feng Xiong
Publications
Correcting What You Cannot See: Credit Assignment for Perception Distillation in Multimodal Reasoners
On-policy distillation provides dense supervision for multimodal reasoners, but its trajectory-level reward cannot determine whether a failed answer arose from perception or subsequent reasoning. Perception Success Rate (PSR), estimated from multiple reasonings sharing one perception, remains ambiguous because low success conflates perceptual insufficiency with reasoning difficulty. We introduce \textbf{Perception-Correction Distillation (PCD)}, a label-free method that identifies correctable perception failures using downstream failure and teacher--student disagreement as complementary witnesses. Their product, , forms a soft AND gate that strengthens distillation only when both witnesses are present. We motivate this rule through Bayesian evidence combination and show that multiplication is the unique normalized bilinear gate that vanishes when either witness is absent. PCD uses separated perception--reasoning rollouts and mean-preserving weights, leaving the reasoning objective unchanged. Across eight benchmarks, PCD improves the 8B 2B macro average from 44.50 with OPD to 47.28 and the 32B 8B result from 56.94 to 61.22. In matched 2B ablations, removing PCD and separated rollout reduces held-out average by 2.22 and 0.88 points, respectively. Effective multimodal distillation therefore depends not only on what the teacher predicts, but also on identifying when perception is the appropriate target of correction.
Visually-Guided Policy Optimization for Multimodal Reasoning
Reinforcement learning with verifiable rewards (RLVR) has significantly advanced the reasoning ability of vision-language models (VLMs). However, the inherent text-dominated nature of VLMs often leads to insufficient visual faithfulness, characterized by sparse attention activation to visual tokens. More importantly, our empirical analysis reveals that temporal visual forgetting along reasoning steps exacerbates this deficiency. To bridge this gap, we propose Visually-Guided Policy Optimization (VGPO), a novel framework to reinforce visual focus during policy optimization. Specifically, VGPO initially introduces a Visual Attention Compensation mechanism that leverages visual similarity to localize and amplify visual cues, while progressively elevating visual expectations in later steps to counteract visual forgetting. Building on this mechanism, we implement a dual-grained advantage re-weighting strategy: the intra-trajectory level highlights tokens exhibiting relatively high visual activation, while the inter-trajectory level prioritizes trajectories demonstrating superior visual accumulation. Extensive experiments demonstrate that VGPO achieves better visual activation and superior performance in mathematical multimodal reasoning and visual-dependent tasks.
MMKG-RDS: Reasoning Data Synthesis via Deep Mining of Multimodal Knowledge Graphs
Synthesizing high-quality training data is crucial for enhancing domain models' reasoning abilities. Existing methods face limitations in long-tail knowledge coverage, effectiveness verification, and interpretability. Knowledge-graph-based approaches still fall short in functionality, granularity, customizability, and evaluation. To address these issues, we propose MMKG-RDS, a flexible framework for reasoning data synthesis that leverages multimodal knowledge graphs. It supports fine-grained knowledge extraction, customizable path sampling, and multidimensional data quality scoring. We validate MMKG-RDS with the MMKG-RDS-Bench dataset, covering five domains, 17 task types, and 14,950 samples. Experimental results show fine-tuning Qwen3 models (0.6B/8B/32B) on a small number of synthesized samples improves reasoning accuracy by 9.2%. The framework also generates distinct data, challenging existing models on tasks involving tables and formulas, useful for complex benchmark construction. The dataset and code are available at https://github.com/360AILAB-NLP/MMKG-RDS