Shapley Interpretation and Activation in Neural Networks

Yadong Li, Xin Cui

We propose a novel Shapley value approach to help address neural networks' interpretability and "vanishing gradient" problems. Our method is based on an accurate analytical approxi…