AdaSTE: An Adaptive Straight-Through Estimator To Train Binary Neural Networks

2022년 09월 02일 (금) | 발표자: 윤태현

In this paper, authors show that AdaSTE, a variant of original Straight Through Estimator, where it improves the performance of the Binary Neural Network with theoretical justifications. Also they provide design choices of AdaSTE on theoretical gorunds, followed by empirical results on CIFAR10/100, and TinyImagenet, outperforming existing Binary Neural Network training methods.