Scaling Up Exact Neural Network Compression by ReLU Stability

Thiago Serra
Bucknell University

We can compress a neural network while exactly preserving its underlying functionality with respect to a given input domain if some of its neurons are stable. However, current approaches to determine the stability of neurons require solving or finding a good approximation to multiple discrete optimization problems. In this talk, we present an algorithm based on solving a single optimization problem to identify all stable neurons. Our approach is on median 21 times faster than the state-of-art method, which allows us to explore exact compression on deeper (5 x 100) and wider (2 x 800) networks within minutes. For classifiers trained under an amount of L1 regularization that does not worsen accuracy, we can remove up to 40% of the connections.

This talk is based on joint work with Srikumar Ramalingam (Google Research) and Abhinav Kumar (Michigan State University).

Presentation (PDF File)

Back to Long Programs