Interpretable Companions for Black-Box Models

Danqing Pan, Tong Wang, Satoshi Hara

We present an interpretable companion model for any pre-trained black-box classifiers. The idea is that for any input, a user can decide to either receive a prediction from the bla…