The Propensity for Density in Feed-forward Models

Nandi Schoots, Alex Jackson, Ali Kholmovaia +2

Does the process of training a neural network to solve a task tend to use all of the available weights even when the task could be solved with fewer weights? To address this questi…