Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Wang, Jiahao, Xu, Weiye, Yang, Aijun, Zhou, Wengang, Lu, Lewei, Li, Houqiang, Wang, Xiaohua, Zhu, Jinguo
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2511.10648
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909900755435520
author Wang, Jiahao
Xu, Weiye
Yang, Aijun
Zhou, Wengang
Lu, Lewei
Li, Houqiang
Wang, Xiaohua
Zhu, Jinguo
author_facet Wang, Jiahao
Xu, Weiye
Yang, Aijun
Zhou, Wengang
Lu, Lewei
Li, Houqiang
Wang, Xiaohua
Zhu, Jinguo
contents Outcome-reward reinforcement learning (RL) is a common and increasingly significant way to refine the step-by-step reasoning of multimodal large language models (MLLMs). In the multiple-choice setting - a dominant format for multimodal reasoning benchmarks - the paradigm faces a significant yet often overlooked obstacle: unfaithful trajectories that guess the correct option after a faulty chain of thought receive the same reward as genuine reasoning, which is a flaw that cannot be ignored. We propose Self-Consistency Sampling (SCS) to correct this issue. For each question, SCS (i) introduces small visual perturbations and (ii) performs repeated truncation and resampling of an initial trajectory; agreement among the resulting trajectories yields a differentiable consistency score that down-weights unreliable traces during policy updates. Based on Qwen2.5-VL-7B-Instruct, plugging SCS into RLOO, GRPO, and REINFORCE++ series improves accuracy by up to 7.7 percentage points on six multimodal benchmarks with negligible extra computation. SCS also yields notable gains on both Qwen2.5-VL-3B-Instruct and InternVL3-8B, offering a simple, general remedy for outcome-reward RL in MLLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2511_10648
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Enhancing the Outcome Reward-based RL Training of MLLMs with Self-Consistency Sampling
Wang, Jiahao
Xu, Weiye
Yang, Aijun
Zhou, Wengang
Lu, Lewei
Li, Houqiang
Wang, Xiaohua
Zhu, Jinguo
Computer Vision and Pattern Recognition
Outcome-reward reinforcement learning (RL) is a common and increasingly significant way to refine the step-by-step reasoning of multimodal large language models (MLLMs). In the multiple-choice setting - a dominant format for multimodal reasoning benchmarks - the paradigm faces a significant yet often overlooked obstacle: unfaithful trajectories that guess the correct option after a faulty chain of thought receive the same reward as genuine reasoning, which is a flaw that cannot be ignored. We propose Self-Consistency Sampling (SCS) to correct this issue. For each question, SCS (i) introduces small visual perturbations and (ii) performs repeated truncation and resampling of an initial trajectory; agreement among the resulting trajectories yields a differentiable consistency score that down-weights unreliable traces during policy updates. Based on Qwen2.5-VL-7B-Instruct, plugging SCS into RLOO, GRPO, and REINFORCE++ series improves accuracy by up to 7.7 percentage points on six multimodal benchmarks with negligible extra computation. SCS also yields notable gains on both Qwen2.5-VL-3B-Instruct and InternVL3-8B, offering a simple, general remedy for outcome-reward RL in MLLMs.
title Enhancing the Outcome Reward-based RL Training of MLLMs with Self-Consistency Sampling
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2511.10648