Provable Guarantees for Gradient-Based Meta-Learning

Abstract

We study the problem of meta-learning through the lens of online convex optimization, developing a meta-algorithm bridging the gap between popular gradient-based meta-learning and classical regularization-based multi-task transfer methods. Our method is the first to simultaneously satisfy good sample efficiency guarantees in the convex setting, with generalization bounds that improve with task-similarity, while also being computationally scalable to modern deep learning architectures and the many-task setting. Despite its simplicity, the algorithm matches, up to a constant factor, a lower bound on the performance of any such parameter-transfer method under natural task similarity assumptions. We use experiments in both convex and deep learning settings to verify and demonstrate the applicability of our theory.

ICML 2019

References in corpus (3)

• Theoretical Models of Learning to Learn
• Meta-SGD: Learning to Learn Quickly for Few-Shot Learning
• A Theoretical Analysis of Contrastive Unsupervised Representation Learning

Cited by in corpus (7)

• Differentiable Meta-learning Model for Few-shot Semantic Segmentation
• On Localized Discrepancy for Domain Adaptation
• Bayes meets Bernstein at the Meta Level: an Analysis of Fast Rates in Meta-Learning with PAC-Bayes
• Variable-Shot Adaptation for Online Meta-Learning
• System Identification via Meta-Learning in Linear Time-Varying Environments
• Meta-Learning Conjugate Priors for Few-Shot Bayesian Optimization
• Online Bilevel Optimization: Regret Analysis of Online Alternating Gradient Methods