Modular meta-learning

TL;DR

This paper introduces a modular meta-learning approach where neural network modules are learned and combined to efficiently adapt to new tasks, demonstrating improved performance in robotics applications.

Contribution

It proposes a novel modular meta-learning framework that enables combinatorial generalization by reusing learned modules across related tasks.

Findings

Achieves better generalization to new tasks through module composition

Demonstrates improved performance on robotics problems

Shows the effectiveness of modular reuse in meta-learning

Abstract

Many prediction problems, such as those that arise in the context of robotics, have a simplifying underlying structure that, if known, could accelerate learning. In this paper, we present a strategy for learning a set of neural network modules that can be combined in different ways. We train different modular structures on a set of related tasks and generalize to new tasks by composing the learned modules in new ways. By reusing modules to generalize we achieve combinatorial generalization, akin to the "infinite use of finite means" displayed in language. Finally, we show this improves performance in two robotics-related problems.