Learned Optimizers that Scale and Generalize

TL;DR

This paper presents a scalable, generalizable learned optimizer based on a hierarchical RNN architecture that outperforms traditional optimizers and generalizes across diverse tasks, including large neural networks on ImageNet.

Contribution

Introduces a novel hierarchical RNN-based learned optimizer that scales, reduces overhead, and generalizes well to unseen tasks and large-scale neural networks.

Findings

Outperforms RMSProp/ADAM on diverse tasks

Generalizes to unseen neural network architectures

Successfully trains large models on ImageNet

Abstract

Learning to learn has emerged as an important direction for achieving artificial intelligence. Two of the primary barriers to its adoption are an inability to scale to larger problems and a limited ability to generalize to new tasks. We introduce a learned gradient descent optimizer that generalizes well to new tasks, and which has significantly reduced memory and computation overhead. We achieve this by introducing a novel hierarchical RNN architecture, with minimal per-parameter overhead, augmented with additional architectural features that mirror the known structure of optimization tasks. We also develop a meta-training ensemble of small, diverse optimization tasks capturing common properties of loss landscapes. The optimizer learns to outperform RMSProp/ADAM on problems in this corpus. More importantly, it performs comparably or better when applied to small convolutional neural networks, despite seeing no neural networks in its meta-training set. Finally, it generalizes to train Inception V3 and ResNet V2 architectures on the ImageNet dataset for thousands of steps, optimization problems that are of a vastly different scale than those it was trained on. We release an open source implementation of the meta-training algorithm.