VL-SAE: Interpreting and Enhancing Vision-Language Alignment with a Unified Concept Set

Re-assign community