TraceICL: Thread-Resolved Demonstration Curation with Consistency Calibration for Robust Multimodal In-Context Learning
Computer Science > Computer Vision and Pattern Recognition arXiv:2610.12516 (cs) [Submitted on 7 Oct 2026] Title: TraceICL: Thread-Resolved Demonstration Curation with Consistency Calibration for Robust Multimodal In-Context Learning Authors: Yigit Yilmaz, Elena Petrova, Mehmet Kaya, Lucia Rossi, Amir Rahman.
计算机科学 > 计算机视觉与模式识别 arXiv:2610.12516 (cs) [提交于 2026 年 10 月 7 日] 标题：TraceICL：用于稳健多模态上下文学习的线程解析演示策划与一致性校准。作者：Yigit Yilmaz, Elena Petrova, Mehmet Kaya, Lucia Rossi, Amir Rahman。
Abstract: Interleaved multimodal in-context learning (ICL) lets frozen large vision-language models (LVLMs) acquire new tasks from a handful of image-text-answer demonstrations, yet it remains fragile: demonstrations bind to the wrong images, redundant exemplars crowd out informative ones, and noisy answers are copied verbatim into predictions.
摘要：交错式多模态上下文学习（ICL）允许冻结的大型视觉语言模型（LVLM）通过少量图像-文本-答案演示来获取新任务，但它仍然很脆弱：演示会绑定到错误的图像，冗余的示例会挤占信息量大的示例，且嘈杂的答案会被逐字复制到预测中。
We present AVA, a training-free framework that organizes the demonstration pipeline of multimodal ICL into three composable stages. A Thread-Resolved Decomposer (TRD) segments the interleaved context into per-sample threads and re-stitches them with explicit boundaries, restoring image-text binding.
我们提出了 AVA，这是一个无需训练的框架，将多模态 ICL 的演示流程组织为三个可组合的阶段。线程解析分解器（TRD）将交错的上下文分割为每个样本的线程，并用明确的边界重新拼接它们，从而恢复图像与文本的绑定。
An Influence-scored Demonstration Selector (IDS) estimates each candidate’s marginal influence with one extra forward pass, removes redundancy, and orders the survivors by descending value. A Weak-to-Strong Consistency Calibration (WSC) module scores each demonstration by a weak sibling model’s reconstruction confidence and softly reweights its contribution during decoding.
影响评分演示选择器（IDS）通过额外的一次前向传递来估计每个候选者的边际影响，去除冗余，并按价值降序排列剩余的示例。弱到强一致性校准（WSC）模块通过弱同级模型的重构置信度对每个演示进行评分，并在解码过程中对其贡献进行软加权。
On five vision-language benchmarks with interleaved 4-shot prompts, AVA improves the strongest same-setting baseline by +1.7 points on average, with gains of up to +2.4 points on VizWiz, shrinks the require-visual performance gap on ScienceQA-IMG, and retains most of its gain under 30% noisy demonstrations, at a 26% latency overhead that an offline cache amortizes. No gradients are ever computed; the LVLM stays frozen.
在五个带有交错 4-shot 提示的视觉语言基准测试中，AVA 将最强的同类基线平均提高了 1.7 个百分点，在 VizWiz 上增益高达 2.4 个百分点，缩小了 ScienceQA-IMG 上对视觉需求性能的差距，并在 30% 噪声演示下保留了大部分增益，同时仅产生 26% 的延迟开销，且该开销可通过离线缓存摊销。过程中不计算任何梯度；LVLM 保持冻结状态。