Universal Boosts, Specific Suppressors: Sparse Autoencoder Steering of Medical Vision-Language Models 事件

Universal Boosts, Specific Suppressors: Sparse Autoencoder Steering of Medical Vision-Language Models arXiv:2605.24977v1 Announce Type: new Abstract: Medical vision-language models (VLMs) often hallucinate findings when generating chest X-ray reports: they fabricate findings that are not present in the image, miss important ones, or locate them incorrectly. We mitigate this without weight updates by decoding-time residual steering on a per-token sparse autoencoder (SAE) basis: Top-$K$ SAEs on l