Bridging Multi-Task Learning and Meta-Learning: Towards Efficient Training and Effective Adaptation

TL;DR

This paper explores the theoretical and empirical connections between multi-task learning and gradient-based meta-learning, showing their similarities and proposing a faster, efficient alternative for few-shot image classification.

Contribution

It demonstrates the theoretical equivalence between MTL and a class of GBML algorithms and introduces a first-order MTL method that is computationally efficient and effective for fast task adaptation.

Findings

MTL shares the same optimization formulation with certain GBML algorithms.

For over-parameterized neural networks, MTL and GBML produce similar predictions on unseen tasks.

The proposed MTL method is significantly faster than traditional GBML on large-scale datasets.

Abstract

Multi-task learning (MTL) aims to improve the generalization of several related tasks by learning them jointly. As a comparison, in addition to the joint training scheme, modern meta-learning allows unseen tasks with limited labels during the test phase, in the hope of fast adaptation over them. Despite the subtle difference between MTL and meta-learning in the problem formulation, both learning paradigms share the same insight that the shared structure between existing training tasks could lead to better generalization and adaptation. In this paper, we take one important step further to understand the close connection between these two learning paradigms, through both theoretical analysis and empirical investigation. Theoretically, we first demonstrate that MTL shares the same optimization formulation with a class of gradient-based meta-learning (GBML) algorithms. We then prove that for over-parameterized neural networks with sufficient depth, the learned predictive functions of MTL and GBML are close. In particular, this result implies that the predictions given by these two models are similar over the same unseen task. Empirically, we corroborate our theoretical findings by showing that, with proper implementation, MTL is competitive against state-of-the-art GBML algorithms on a set of few-shot image classification benchmarks. Since existing GBML algorithms often involve costly second-order bi-level optimization, our first-order MTL method is an order of magnitude faster on large-scale datasets such as mini-ImageNet. We believe this work could help bridge the gap between these two learning paradigms, and provide a computationally efficient alternative to GBML that also supports fast task adaptation.