Learning to Learn by Zeroth-Order Oracle

Yangjun Ruan, Yuanhao Xiong, Sashank Reddi, Sanjiv Kumar, Cho-Jui Hsieh

Keywords: learning to learn, zeroth-order optimization, black-box adversarial attack

TL;DR: Novel variant of learning to learn framework for zeroth-order optimization that learns both the update rule and the Gaussian sampling rule.

Abstract: In the learning to learn (L2L) framework, we cast the design of optimization algorithms as a machine learning problem and use deep neural networks to learn the update rules. In this paper, we extend the L2L framework to zeroth-order (ZO) optimization setting, where no explicit gradient information is available. Our learned optimizer, modeled as recurrent neural network (RNN), first approximates gradient by ZO gradient estimator and then produces parameter update utilizing the knowledge of previous iterations. To reduce high variance effect due to ZO gradient estimator, we further introduce another RNN to learn the Gaussian sampling rule and dynamically guide the query direction sampling. Our learned optimizer outperforms hand-designed algorithms in terms of convergence rate and final solution on both synthetic and practical ZO optimization tasks (in particular, the black-box adversarial attack task, which is one of the most widely used tasks of ZO optimization). We finally conduct extensive analytical experiments to demonstrate the effectiveness of our proposed optimizer.

Code: https://github.com/RYoungJ/ZO-L2L

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:1910.09464/code)