Automated Curriculum Learning for Neural Networks

TL;DR

This paper presents an automated curriculum learning method that dynamically selects training paths for neural networks using a bandit algorithm, significantly improving learning efficiency and reducing training time.

Contribution

It introduces a novel automatic curriculum selection method using a bandit algorithm guided by learning progress signals, enhancing neural network training efficiency.

Findings

Accelerates learning, sometimes halving training time.

Effective across different signals of learning progress.

Applicable to LSTM networks on multiple curricula.

Abstract

We introduce a method for automatically selecting the path, or syllabus, that a neural network follows through a curriculum so as to maximise learning efficiency. A measure of the amount that the network learns from each data sample is provided as a reward signal to a nonstationary multi-armed bandit algorithm, which then determines a stochastic syllabus. We consider a range of signals derived from two distinct indicators of learning progress: rate of increase in prediction accuracy, and rate of increase in network complexity. Experimental results for LSTM networks on three curricula demonstrate that our approach can significantly accelerate learning, in some cases halving the time required to attain a satisfactory performance level.