Learning to Transfer Learn: Reinforcement Learning-Based Selection for Adaptive Transfer Learning

TL;DR

This paper introduces L2TL, an adaptive transfer learning framework that uses reinforcement learning to optimize the transfer process, significantly improving performance especially on small or mismatched datasets.

Contribution

The paper presents a novel RL-guided adaptive transfer learning method that optimizes shared weights and loss weights for better transfer performance.

Findings

L2TL outperforms fine-tuning baselines on eight datasets.

L2TL is especially effective with small-scale and mismatched datasets.

Adaptive weight adjustment improves transfer learning results.

Abstract

We propose a novel adaptive transfer learning framework, learning to transfer learn (L2TL), to improve performance on a target dataset by careful extraction of the related information from a source dataset. Our framework considers cooperative optimization of shared weights between models for source and target tasks, and adjusts the constituent loss weights adaptively. The adaptation of the weights is based on a reinforcement learning (RL) selection policy, guided with a performance metric on the target validation set. We demonstrate that L2TL outperforms fine-tuning baselines and other adaptive transfer learning methods on eight datasets. In the regimes of small-scale target datasets and significant label mismatch between source and target datasets, L2TL shows particularly large benefits.