Less is More: Efficient Black-box Attribution via Minimal Interpretable Subset Selection

ArXi:2504.00470v2 Announce Type: replace To develop a trustworthy AI system, which aim to identify the input regions that most influence the models decisions. The primary task of existing attribution methods lies in efficiently and accurately identifying the relationships among input-prediction interactions. Particularly when the input data is discrete, such as images, analyzing the relationship between inputs and outputs poses a significant challenge due to the combinatorial explosion.