| Paper ID | D5-S5-T3.2 |
| Paper Title |
Information-theoretic lower bounds for zero-order stochastic gradient estimation |
| Authors |
Abdulrahman Alabdulkareem, Jean Honorio, Purdue University, United States |
| Session |
D5-S5-T3: Optimization |
| Chaired Session: |
Friday, 16 July, 23:20 - 23:40 |
| Engagement Session: |
Friday, 16 July, 23:40 - 00:00 |
| Abstract |
In this paper we analyze the necessary number of samples to estimate the gradient of any multidimensional smooth (possibly non-convex) function in a zero-order stochastic oracle model. In this model, an estimator has access to noisy values of the function, in order to produce the estimate of the gradient. We also provide an analysis on the sufficient number of samples for the finite difference method, a classical technique in numerical linear algebra. For $T$ samples and $d$ dimensions, our information-theoretic lower bound is $\Omega(\sqrt{d/T})$. We show that the finite difference method for a bounded-variance oracle has rate $O(d^{4/3}/\sqrt{T})$ for functions with zero third and higher order derivatives. These rates are tight for Gaussian oracles. Thus, the finite difference method is not minimax optimal, and therefore there is space for the development of better gradient estimation methods.
|
