Multimodal Model Diffing for Feature Discovery and Control
论文概要
研究领域: NLP 作者: Hunar Batra, Lachin Naghashyar, Ashkan Khakzar 发布时间: 2026-08-11 arXiv: 2508.03805
中文摘要
多模态大语言模型(MLLMs)展现出强大的视觉理解能力,但导致这些行为的内部特征仍然难以识别、审计或控制。虽然稀疏自编码器(SAEs)将隐藏状态分解为可解释的特征方向适用于事后检查,但它们既不能轻易隔离哪些特征被多模态训练改变,也不能直接用于有针对性的控制。我们引入了MMDiff,一个多模态模型差异分析框架,训练多模态SAE并将其转变为用于发现和控制多模态行为的特征级接口。MMDiff支持三种用途:(i) 特征隔离,通过对比基础LM SAE与其多模态适应版本来识别被多模态训练改变的特征;(ii) 任务特定特征检测,通过每token对比触发分析来隔离因果特征;(iii) 特征级控制,通过因果移除或引导发现的特征方向。我们为三个MLLM家族(LLaVA-MORE、PaliGemma 2和InternVL3.5)训练了多模态SAE,并在视觉空间理解、多模态安全性和OCR上进行了评估。MMDiff发现了稀疏的、因果特定的特征,其移除在空间任务上有选择性地降低目标行为12%,在OCR上降低17%,在多模态安全攻击中降低攻击成功率24%,且不影响VQA性能。引导这些特征比标准单层引导基线在空间和OCR准确性上平均提高+3.6%和+1.8%。这些结果表明,多模态SAE不仅可以作为可解释性工具,还可以作为审计、引导和控制MLLM行为以实现更安全、更有能力的生成的机制。
原文摘要
Multimodal Large Language Models (MLLMs) exhibit strong visual understanding, yet the internal features that cause these behaviors remain difficult to identify, audit, or control. While applicable to post-hoc inspection, hidden states that are decomposed into interpretable feature directions using sparse autoencoders (SAEs) neither readily isolate which features are changed by multimodal training, nor are they directly useful for targeted control. We introduce MMDiff, a multimodal model-diffing framework that trains multimodal SAEs and turns them into feature-level interfaces for discovering and controlling multimodal behavior. MMDiff supports three uses: (i) feature isolation, by diffing a base-LM SAE against its multimodal-adapted counterpart to identify features altered by multimodal tr...
--- *自动采集于 2026-08-12*
#论文 #arXiv #NLP #小凯