Abstract
Recently authors have introduced the idea of training discrete weights neural networks using a mix between classical simulated annealing and a replica ansatz known from the statistical physics literature. Among other points, they claim their method is able to find robust configurations. In this paper, we analyze this so called “replicated simulated annealing” algorithm. In particular, we give criteria to guarantee its convergence, and study when it successfully samples from configurations. We also perform experiments using synthetic and real data bases.
Highlights
In the past few years, there has been a growing interest in finding methods to train discrete weights neural networks
The outline is as follows: in Sect. 2 we mathematically formalize and describe the algorithm of Replicated Simulated Annealing, in Sect. 3 we study its convergence properties
When training a binary logistic regression classifier on MNIST using Replicated Simulated Annealing, we typically achieve a 88% accuracy on the test set, which is on par with the performance obtained with continuous weights and gradient descent
Summary
In the past few years, there has been a growing interest in finding methods to train discrete weights neural networks. As a matter of fact, when it comes to implementations, Communicated by Hal Tasaki.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have