1 paper · 1 filter
Li Puyin, Jiyuan Tan, Ahmad Jabbar +2
We present a method for diagnosing interpretation in neural networks by identifying an input subspace where a proposed interpretation is highly faithful. Our method is particularly…