Seeing Before Reasoning: Decoupling Perception and Reasoning for Shortcut-Resilient Multimodal On-Policy Self-Distillation
先看后思:解耦感知与推理以实现抗捷径的多模态在策略自蒸馏
机构 * State Key Laboratory of Robotics and Intelligent Systems, Shenyang Institute of Automation, Chinese Academy of Sciences(机器人与智能系统国家重点实验室,沈阳自动化研究所,中国科学院) ; University of Chinese Academy of Sciences(中国科学院大学)
专题命中 推理与问题求解 :LLM(abstract,abstract_cn);large language model(abstract);language model(abstract);post-training(abstract)
AI总结 提出ViGOS框架,通过解耦感知和推理,在MLLM后训练中避免文本捷径,提升图像依赖行为。
Comments 29 pages, 5 figures, 8 tables; Project page: https://oedosoldier.github.io/ViGOS/