10
53

Time/Accuracy Tradeoffs for Learning a ReLU with respect to Gaussian Marginals

Abstract

We consider the problem of computing the best-fitting ReLU with respect to square-loss on a training set when the examples have been drawn according to a spherical Gaussian distribution (the labels can be arbitrary). Let opt<1\mathsf{opt} < 1 be the population loss of the best-fitting ReLU. We prove: 1. Finding a ReLU with square-loss opt+ϵ\mathsf{opt} + \epsilon is as hard as the problem of learning sparse parities with noise, widely thought to be computationally intractable. This is the first hardness result for learning a ReLU with respect to Gaussian marginals, and our results imply -{\emph unconditionally}- that gradient descent cannot converge to the global minimum in polynomial time. 2. There exists an efficient approximation algorithm for finding the best-fitting ReLU that achieves error O(opt2/3)O(\mathsf{opt}^{2/3}). The algorithm uses a novel reduction to noisy halfspace learning with respect to 0/10/1 loss. Prior work due to Soltanolkotabi [Sol17] showed that gradient descent can find the best-fitting ReLU with respect to Gaussian marginals, if the training set is exactly labeled by a ReLU.

View on arXiv
Comments on this paper