Reinforcement learning-powered co-optimization: Bridging critic model and multimodal LLM reasoning abilities
Authors:
- Qing WangQWQing Wang
- Shuhang LiuSLShuhang Liu
- Zhenrong ZhangZZZhenrong Zhang
- Jun DuJDJun Du
- Jianshu ZhangJZJianshu Zhang
- Quan LiuQLQuan Liu
- 15 May 2026
- journal article
- Published by Elsevier in Pattern Recognition
- Vol. 180
Abstract
No abstract availableFunding Information
- National Natural Science Foundation of China (U25A20409)
This publication has 17 references indexed in Scilit:
- KAID: Knowledge-Aware Interactive Distillation for Vision-Language Models2025
- SVGen: Interpretable Vector Graphics Generation with Large Language Models2025
- MMC: Iterative Refinement of VLM Reasoning via MCTS-based Multimodal Critique2025
- R1-Onevision: Advancing Generalized Multimodal Reasoning Through Cross-Modal Formalization2025
- LlaVA-CoT: Let Vision Language Models Reason Step-By-Step2025
- DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learningNature, 2025
- Critic-V: VLM Critics Help Catch VLM Errors in Multimodal Reasoning2025
- MMMU: A Massive Multi-Discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI2024
- InfographicVQA2022
- DocVQA: A Dataset for VQA on Document Images2021