Publication Date
2021
Conference/Sponsorship/Institution
Thirty-fifth Conference on Neural Information Processing Systems (NeurIPS 2021)
Description
We can compress a rectifier 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 with Rectified Linear Unit (ReLU) activations require solving or finding a good approximation to multiple discrete optimization problems. In this work, we introduce an algorithm based on solving a single optimization problem to identify all stable neurons. Our approach is on median 183 times faster than the state-of-art method on CIFAR-10, 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 56% of the connections on the CIFAR-10 dataset. The code is available at the following link, https://github.com/yuxwind/ExactCompression .
Type
Conference Paper
Department
Analytics & Operations Management
Link to published version
https://arxiv.org/pdf/2102.07804.pdf
Recommended Citation
Serra, Thiago; Yu, Xin; Kumar, Abhinav; and Ramalingam, Srikumar, "Scaling Up Exact Neural Network Compression by ReLU Stability" (2021). Faculty Conference Papers and Presentations. 62.
https://digitalcommons.bucknell.edu/fac_conf/62
Included in
Artificial Intelligence and Robotics Commons, Other Applied Mathematics Commons, Theory and Algorithms Commons