Transferring Knowledge across Learning Processes

TL;DR

Leap is a novel transfer learning framework that transfers knowledge across entire learning processes by minimizing the path length on task manifolds, improving performance in vision and reinforcement learning tasks.

Contribution

The paper introduces Leap, a framework that transfers knowledge across learning processes using path length minimization on task manifolds, a higher-level abstraction than parameter transfer.

Findings

Leap outperforms existing methods in meta-learning and transfer learning on vision tasks.

Leap effectively transfers knowledge in complex reinforcement learning environments.

The framework is computationally efficient, using only training information.

Abstract

In complex transfer learning scenarios new tasks might not be tightly linked to previous tasks. Approaches that transfer information contained only in the final parameters of a source model will therefore struggle. Instead, transfer learning at a higher level of abstraction is needed. We propose Leap, a framework that achieves this by transferring knowledge across learning processes. We associate each task with a manifold on which the training process travels from initialization to final parameters and construct a meta-learning objective that minimizes the expected length of this path. Our framework leverages only information obtained during training and can be computed on the fly at negligible cost. We demonstrate that our framework outperforms competing methods, both in meta-learning and transfer learning, on a set of computer vision tasks. Finally, we demonstrate that Leap can transfer knowledge across learning processes in demanding reinforcement learning environments (Atari) that involve millions of gradient steps.