Opening the black box of neural nets: case studies in stop/top discrimination

Thomas Roxlo, Matthew Reece

We introduce techniques for exploring the functionality of a neural network and extracting simple, human-readable approximations to its performance. By performing gradient ascent o…