EvoGrad: Efficient Gradient-Based Meta-Learning and Hyperparameter Optimization

TL;DR

EvoGrad introduces an evolutionary approach to hypergradient computation that enhances efficiency and scalability in gradient-based meta-learning and hyperparameter optimization, avoiding costly second-order derivatives.

Contribution

EvoGrad proposes a novel evolutionary method for hypergradient estimation that reduces computational costs and enables scaling to larger neural network architectures.

Findings

EvoGrad significantly improves efficiency over traditional methods.

EvoGrad enables scaling meta-learning to larger architectures like ResNet34.

EvoGrad performs well across diverse meta-learning applications.

Abstract

Gradient-based meta-learning and hyperparameter optimization have seen significant progress recently, enabling practical end-to-end training of neural networks together with many hyperparameters. Nevertheless, existing approaches are relatively expensive as they need to compute second-order derivatives and store a longer computational graph. This cost prevents scaling them to larger network architectures. We present EvoGrad, a new approach to meta-learning that draws upon evolutionary techniques to more efficiently compute hypergradients. EvoGrad estimates hypergradient with respect to hyperparameters without calculating second-order gradients, or storing a longer computational graph, leading to significant improvements in efficiency. We evaluate EvoGrad on three substantial recent meta-learning applications, namely cross-domain few-shot learning with feature-wise transformations, noisy label learning with Meta-Weight-Net and low-resource cross-lingual learning with meta representation transformation. The results show that EvoGrad significantly improves efficiency and enables scaling meta-learning to bigger architectures such as from ResNet10 to ResNet34.