arXiv · Computer Vision· Yigit Yilmaz, Elena Petrova, Mehmet Kaya, Lucia Rossi, Amir Rahman·Oct 12, 2026, 12:00· 3 hr agoAI score25

TraceICL：线程解析式示例筛选与一致性校准提升多模态上下文学习鲁棒性

TraceICL: Thread-Resolved Demonstration Curation with Consistency Calibration for Robust Multimodal In-Context Learning

AI brief

论文提出免训练框架 AVA，将多模态上下文学习示例管线拆为线程解析分解器（TRD）、影响力评分示例选择器（IDS）与弱到强一致性校准（WSC）三个可组合阶段，修复图文绑定、剔除冗余示例并抑制噪声答案。