Abstract
Reinforcement Learning with Verifiable Rewards (RLVR) has been extended to Large Vision-Language Models (LVLMs), and perception-aware methods further encourage policies to rely on visual evidence. Yet relying on the image does not guarantee that visual claims are supported by it. Before RL training, 27.81% of the correctly answered responses of Qwen2.5-VL-7B on four multimodal reasoning benchmarks contain at least one direct visual claim that the image does not support. Since outcome-level RL rewards each response as a whole, these claims inherit the positive credit of the correct answer. We introduce a fixed-rollout counterfactual diagnostic that re-scores the same response under an intervened image to separate Evidence-Function Sensitivity (EFS), how strongly the model's predictions change, from claim persistence, whether the model keeps supporting the same claim rather than retracting it. The diagnostic reveals Sensitivity-Persistence Decoupling (SPD): under DAPO and VPPO, EFS increases and claims become more retractable overall, yet unsupported claims become significantly more persistent, whereas GRPO raises EFS without this deterioration. We therefore propose Persistence-Aware Credit Gating (PACG), which attenuates positive credit for unusually persistent visual claims and leaves all other credit unchanged. It requires no supported/unsupported labels and adds no inference cost. On Qwen2.5-VL-7B, PACG raises the nine-benchmark average over three seeds from 58.1% to 59.9% with DAPO and from 59.8% to 60.9% with VPPO, while making unsupported claims more retractable. The gains extend to a larger model, a newer backbone, and the accuracy of HallusionBench also improves consistently. These results suggest that visual sensitivity and claim retractability are complementary dimensions of multimodal credit assignment.
Keywords
Subject
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Bi, Z., Lin, K., Gao, X., Sun, Y., & Zhang, L. (2026). Visual sensitivity is not claim retractability: persistence-aware credit assignment for multimodal reinforcement learning. https://omanscience.com/en/articles/visual-sensitivity-is-not-claim-retractability-persistence-aware-credit-assignment-for-multimodal-reinforcement-learning
MLA 9
Bi, Zhongan, et al. "Visual sensitivity is not claim retractability: persistence-aware credit assignment for multimodal reinforcement learning." https://omanscience.com/en/articles/visual-sensitivity-is-not-claim-retractability-persistence-aware-credit-assignment-for-multimodal-reinforcement-learning.
Chicago (author–date)
Bi, Zhongan, Kepeng Lin, Xuanang Gao, Yuhan Sun, and Lianrun Zhang. 2026. "Visual sensitivity is not claim retractability: persistence-aware credit assignment for multimodal reinforcement learning." https://omanscience.com/en/articles/visual-sensitivity-is-not-claim-retractability-persistence-aware-credit-assignment-for-multimodal-reinforcement-learning.
Harvard
Bi, Z., Lin, K., Gao, X., Sun, Y. and Zhang, L. (2026) 'Visual sensitivity is not claim retractability: persistence-aware credit assignment for multimodal reinforcement learning', Available at: https://omanscience.com/en/articles/visual-sensitivity-is-not-claim-retractability-persistence-aware-credit-assignment-for-multimodal-reinforcement-learning.
Vancouver
Bi Z, Lin K, Gao X, Sun Y, Zhang L. Visual sensitivity is not claim retractability: persistence-aware credit assignment for multimodal reinforcement learning. https://omanscience.com/en/articles/visual-sensitivity-is-not-claim-retractability-persistence-aware-credit-assignment-for-multimodal-reinforcement-learning
IEEE
Z. Bi, K. Lin, X. Gao, Y. Sun, and L. Zhang, "Visual sensitivity is not claim retractability: persistence-aware credit assignment for multimodal reinforcement learning," https://omanscience.com/en/articles/visual-sensitivity-is-not-claim-retractability-persistence-aware-credit-assignment-for-multimodal-reinforcement-learning.