Detecting Statistical Interactions from Neural Network Weights

We develop a method of detecting statistical interactions in data by directly interpreting the trained weights of a feedforward multilayer neural network. With regularization applied to the weights, our method can achieve similar interaction detection performance compared to the state-of-the-art without searching an exponential solution space of possible interactions. We obtain our computational savings by first observing that interactions between input features are created by the non-additive effect of nonlinear activation functions, and that interacting paths are encoded in weight matrices. We use these observations to develop a way of identifying interactions without assuming their order or functional form via a simple traversal over the input weight matrix. The generality of these interactions provides simultaneous insight into the complex functions within feedforward networks and data. In experiments, we demonstrate the performance of our method and the importance of discovered interactions.