paper-with-me

Papers

Thinking with Deltas: Incentivizing Reinforcement Learning via Differential Visual Reasoning Policy

2026-01-11 · Shujian Gao, Yuan Wang, Jiangtao Yan, Zuxuan Wu, Yu-Gang Jiang arxiv

Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced reasoning capabilities in Large Language Models. However, adapting RLVR to multimodal domains suffers from a critical \textit{perception-reasoning decoupling}. Existing paradigms, driven by text-centric outcome rewards, reasoning in language medium, inadvertently encourage models to bypass visual perception. We empirically validate this through blind experiments: state-of-the-art policies maintain or surprisingly improve performance even when visual inputs are entirely removed. This reveals that these models degenerate into \textit{blind reasoners}, exploiting linguistic priors to generate plausible answers instead of attending to visual evidence. In response, we propose \textbf{Thinking with Deltas}, a framework driven by a \textbf{Differential Visual Reasoning Policy (DVRP)}. DVRP introduces intrinsic supervision via visual triplets, comprising original, masked, and perturbed inputs. It optimizes the model to maximize reasoning divergence from masked inputs (enforcing \textit{visual sensitivity}) while minimizing divergence from perturbed inputs (ensuring \textit{visual robustness}). By aligning reasoning variations strictly with the \textit{Delta} of visual information, DVRP inherently bolsters visual understanding capabilities and significantly outperforms state-of-the-art methods on both general and medical benchmarks, without requiring external annotations or auxiliary tools.

📄 PDF Abstract BibTeX arXiv:2601.06801

Code (0)

등록된 구현이 없습니다.

Tasks

Reinforcement LearningVisual Reasoning

Similar Papers 제목 키워드 기반

DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning

2025-05-20 · Ziwei Zheng, Michael Yang, Jack Hong, Chenxiao Zhao 외

Large Vision-Language Models (VLMs) have shown strong capabilities in multimodal understanding and reasoning, yet they are primarily constrained by text-based reasoning processes. However, achieving seamless integration …

HallucinationMathematical ReasoningMultimodal Reasoningreinforcement-learning+2

VisualDeltas: Learning Preferences from Visual Quality Perturbations

2026-03-07 · Hailiang Huang, Yihao Liu, Shengyue Guan, Haoze Li 외 arxiv

We present VisualDeltas, a lightweight preference-learning framework that extracts supervision from visual quality variations in multimodal data. By leveraging the systematic impact of image quality on visual perception …

VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning

2025-04-10 · Haozhe Wang, Chao Qu, Zuming Huang, Wei Chu 외

Recently, slow-thinking systems like GPT-o1 and DeepSeek-R1 have demonstrated great potential in solving challenging problems through explicit reflection. They significantly outperform the best fast-thinking models, such…

MathMultimodal Reasoning

Incentivizing Dual Process Thinking for Efficient Large Language Model Reasoning

2025-05-22 · Xiaoxue Cheng, Junyi Li, Zhenduo Zhang, Xinyu Tang 외

Large reasoning models (LRMs) have demonstrated strong performance on complex reasoning tasks, but often suffer from overthinking, generating redundant content regardless of task difficulty. Inspired by the dual process …

Language ModelingLanguage ModellingLarge Language Modelreinforcement-learning+1

J1: Incentivizing Thinking in LLM-as-a-Judge via Reinforcement Learning

2025-05-15 · Chenxi Whitehouse, Tianlu Wang, Ping Yu, Xian Li 외

The progress of AI is bottlenecked by the quality of evaluation, and powerful LLM-as-a-Judge models have proved to be a core solution. Improved judgment ability is enabled by stronger chain-of-thought reasoning, motivati…