Generalization Bounds for Neural Networks via Approximate Description Length

Abstract

We investigate the sample complexity of networks with bounds on the magnitude of its weights. In particular, we consider the class \[ H= \W_t W_1 :W_1, ,W_t-1 M_d, d, W_t M_1,d \ \] where the spectral norm of each W_i is bounded by O(1), the Frobenius norm is bounded by R, and is the sigmoid function e^x1+e^x or the smoothened ReLU function (1+e^x). We show that for any depth t, if the inputs are in [-1,1]^d, the sample complexity of H is O(dR^2^2). This bound is optimal up to log-factors, and substantially improves over the previous state of the art of O(d^2R^2^2). We furthermore show that this bound remains valid if instead of considering the magnitude of the W_i's, we consider the magnitude of W_i - W_i^0, where W_i^0 are some reference matrices, with spectral norm of O(1). By taking the W_i^0 to be the matrices at the onset of the training process, we get sample complexity bounds that are sub-linear in the number of parameters, in many typical regimes of parameters. To establish our results we develop a new technique to analyze the sample complexity of families H of predictors. We start by defining a new notion of a randomized approximate description of functions f:XR^d. We then show that if there is a way to approximately describe functions in a class H using d bits, then d/^2 examples suffices to guarantee uniform convergence. Namely, that the empirical loss of all the functions in the class is -close to the true loss. Finally, we develop a set of tools for calculating the approximate description length of classes of functions that can be presented as a composition of linear function classes and non-linear functions.

Tasks

Reproductions