arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2503.22720cs.LGcs.AI
收藏

为何表征工程有效:基于视觉-语言模型的理论与实证研究

Why Representation Engineering Works: A Theoretical and Empirical Study in Vision-Language Models

University of Maryland(马里兰大学);North Carolina State University(北卡罗来纳州立大学);Rutgers University(罗格斯大学)

Bowei Tian, Xuntao Lyu, Meng Liu, Hongyi Wang, Ang Li

AI总结:

本研究首次将表征工程扩展至视觉-语言模型以应对视觉输入引发的幻觉问题,提出基于主特征向量的理论框架揭示RepE底层机制,并将其转化为结构化理论框架以提升AI鲁棒性、公平性与透明性。

摘要:

表征工程(Representation Engineering, RepE)已成为一种强大的范式,它通过关注高层表征而非单个神经元或回路来增强AI的透明性。它已被证明在提升可解释性和控制力方面非常有效,表明表征可以在大型语言模型(LLMs)中涌现、传播并塑造最终的模型输出。然而,在视觉-语言模型(VLMs)中,视觉输入可能会覆盖事实性语言知识,导致产生与现实相矛盾的幻觉响应。为了应对这一挑战,我们首次尝试将RepE扩展至VLMs,分析多模态表征如何被保留和转换。基于我们的发现并从成功的RepE应用中汲取灵感,我们开发了一个理论框架,该框架利用主特征向量解释了跨层神经活动的稳定性,揭示了RepE的底层机制。我们实证验证了这些内在属性,证明了它们的广泛适用性和重要性。通过将理论洞察与实证验证相结合,这项工作将RepE从一种描述性工具转化为结构化的理论框架,为提升AI的鲁棒性、公平性和透明性开辟了新方向。

英文摘要:

Representation Engineering (RepE) has emerged as a powerful paradigm for enhancing AI transparency by focusing on high-level representations rather than individual neurons or circuits. It has proven effective in improving interpretability and control, showing that representations can emerge, propagate, and shape final model outputs in large language models (LLMs). However, in Vision-Language Models (VLMs), visual input can override factual linguistic knowledge, leading to hallucinated responses that contradict reality. To address this challenge, we make the first attempt to extend RepE to VLMs, analyzing how multimodal representations are preserved and transformed. Building on our findings and drawing inspiration from successful RepE applications, we develop a theoretical framework that explains the stability of neural activity across layers using the principal eigenvector, uncovering the underlying mechanism of RepE. We empirically validate these instrinsic properties, demonstrating their broad applicability and significance. By bridging theoretical insights with empirical validation, this work transforms RepE from a descriptive tool into a structured theoretical framework, opening new directions for improving AI robustness, fairness, and transparency.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑