Skip to content
Preprint

Grounding Multimodal Reasoning with Evidence-Ablated Negatives

Oct 2026 · 0 citations · 64 references
Computer Science

TL;DR

EVAN is proposed, a multimodal reinforcement learning framework utilizing EVidence-Ablated Negatives to train LMMs against their own language-prior failures and mitigates modality imbalance and improves over strong RLVR baselines.

Abstract

Reinforcement learning with verifiable rewards (RLVR) has become a powerful tool for improving multimodal reasoning. However, standard outcome supervision suffers from credit assignment ambiguity, as a model may receive positive rewards by exploiting language priors or dataset shortcuts rather than grounding its reasoning in the requisite visual evidence. This makes high-quality negative responses critical, since effective negatives should remain plausible under the current policy while exposing the shortcut that should be suppressed. In this paper, we propose EVAN, a multimodal reinforcement learning framework utilizing EVidence-Ablated Negatives to train LMMs against their own language-prior failures. EVAN first identifies highly visually dependent examples using a scaled image information gain metric and preserves data diversity with a mixed-ranking strategy. It then constructs evidence-ablated images by masking the most discriminative image patches, eliciting plausible but visually ungrounded responses from the current policy. During policy optimization, these negative responses are incorporated as bounded auxiliary signals under the clean image, penalizing shortcut learning while maintaining RL stability. Experiments across challenging multimodal reasoning benchmarks demonstrate that EVAN mitigates modality imbalance and improves over strong RLVR baselines. Our code is available at https://github.com/DawnLIU35/EVAN.git.

View source

Similar papers

Preprint Nov 2026

MIRA: Reinforcing Multimodal Reasoning via Deceptive Contextual Augmentation

While Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced multimodal reasoning, existing frameworks suffer from a critical"perception-reasoning gap."Due to an over-reliance on seemingly relevant visual tokens, even minor perceptual perturbations can propagate through the reasoning process,...

Zhi-Han Yin, Jian-Xin Liang, Yi-Feng Yao et al. · 1 citation
#artificial intelligence Preprint Oct 2026

From Visual Affinity to Learning Credit in Multimodal Reinforcement Learning

Reinforcement learning with verifiable rewards (RLVR) improves multimodal reasoning, but uniformly assigning a single response-level advantage to all tokens overlooks the distinct roles of visual observations and subsequent reasoning. Visual affinity alone is insufficient: predictive states may align with both image an...

Yu-Hang Han, Ding-Ning Liu, Yu-Jie Chen et al. · 0 citations
#artificial intelligence Preprint Oct 2026

PerceptRL: Making Visual Perception Rewardable in MLLMs from Answers to Interactions

PerceptRL is presented, which strengthens perception at two complementary levels in multimodal large language models by using a multi-agent pipeline to extract verified errors from dense image captions and convert the corresponding visual facts into QA pairs with fixed answer labels and rule-based rewards.

Tong-Kun Guan, Zhi-Bo Yang, Ming-Kun Yang et al. · 0 citations

Visual sensitivity is not claim retractability: persistence-aware credit assignment for multimodal reinforcement learning

Persistence-Aware Credit Gating (PACG), which attenuates positive credit for unusually persistent visual claims and leaves all other credit unchanged, which extends to a larger model, a newer backbone, and the accuracy of HallusionBench also improves consistently.

Zhong-An Bi, Ke-Peng Lin, Xuan-Ang Gao et al. · 0 citations
#artificial intelligence Preprint Oct 2026

GRIT: Gradient Masking for Visually Misgrounded Tokens in Text-Rich Reasoning

GRIT, a training framework that incorporates visual grounding into RLVR to mitigate the impact of Correct-but-Visually-Misgrounded Reasoning (CVMR), is proposed and introduced, and experiments demonstrate that GRIT significantly improves both reasoning accuracy and grounding fidelity over existing RLVR methods.

Chanseul Cho, No-Chan Bae, Dongheon Lee et al. · 0 citations
Preprint Oct 2026

Reinforcement Learning with Visually Attributable On-Policy Self-Distillation

Multimodal reinforcement learning with verifiable rewards (RLVR) provides reliable task-level supervision, yet offers limited guidance for localizing errors in visual reasoning. On-policy self-distillation (OPSD) complements this supervision with dense token-level corrections from a privileged teacher equipped with an...

Shiao Xie, Si-Yu Chen, Jian-Wei Lv et al. · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.