AI RESEARCH

SALMUBench: A Benchmark for Sensitive Association-Level Multimodal Unlearning

arXiv CS.LG

ArXi:2603.26316v1 Announce Type: cross As multimodal models like CLIP become integral to downstream systems, the need to remove sensitive information is critical. However, machine unlearning for contrastively-trained encoders remains underexplored, and existing evaluations fail to diagnose fine-grained, association-level forgetting. We