Automated Curriculum Learning for Neural Networks
This addresses the challenge of optimizing curriculum learning for neural networks, which is an incremental improvement in training efficiency for machine learning practitioners.
The paper tackles the problem of automatically selecting a curriculum to maximize neural network learning efficiency by using a nonstationary multi-armed bandit algorithm with reward signals based on learning progress indicators. The result shows that this approach can significantly accelerate learning, halving the time to achieve satisfactory performance in some cases.
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.