Curvature-Guided Mixing for MLLM Adaptation
曲率引导混合用于MLLM适配
机构 * Research Institute of Trustworthy Autonomous Systems and Department of Computer Science and Engineering, Southern University of Science and Technology(南方科技大学可信自主系统研究院与计算机科学与工程系) ; Department of Computer Science, City University of Hong Kong(香港城市大学计算机科学系)
专题命中 VLM训练与架构 :MLLM(title,title_cn);LLaVA(summary_cn,abstract);multimodal large language model(abstract);分类 cs.CV、cs.LG
AI总结 提出曲率引导混合(CGM)框架,通过二阶Hessian近似推导最优软混合比,以及硬混合变体CGM†,在LLaVA-1.5和Qwen2.5VL上改善任务专化与通用知识保持的权衡。
Comments Accepted to ECCV 2026