Evaluating Attribution Methods using White-Box LSTMs

Yiding Hao

Interpretability methods for neural networks are difficult to evaluate because we do not understand the black-box models typically used to test them. This paper proposes a framewor…