Interpretability Transfer from Language to Vision via Sparse Autoencoders 事件

Interpretability Transfer from Language to Vision via Sparse Autoencoders arXiv:2605.24946v1 Announce Type: new Abstract: Recent advances in language model interpretability using sparse autoencoders (SAEs) have yet to effectively translate to the visual domain, mainly due to the difficulty and ambiguity of labeling visual concepts. In this paper, we introduce Visual Interpretability via SAE Transfer Alignment (VISTA), a framework that transfers interpretability from language to vision in a LLaV