Statistically Significant Stopping of Neural Network Training
This addresses runtime efficiency for scenarios like hyper-parameter tuning where many networks are trained simultaneously, though it is incremental as it builds on existing stopping criteria.
The authors tackled the problem of determining when to stop training neural networks by introducing a statistical significance test to detect when learning has stopped, achieving comparable accuracy to the best methods in 77% or fewer epochs without significant loss in final accuracy.
The general approach taken when training deep learning classifiers is to save the parameters after every few iterations, train until either a human observer or a simple metric-based heuristic decides the network isn't learning anymore, and then backtrack and pick the saved parameters with the best validation accuracy. Simple methods are used to determine if a neural network isn't learning anymore because, as long as it's well after the optimal values are found, the condition doesn't impact the final accuracy of the model. However from a runtime perspective, this is of great significance to the many cases where numerous neural networks are trained simultaneously (e.g. hyper-parameter tuning). Motivated by this, we introduce a statistical significance test to determine if a neural network has stopped learning. This stopping criterion appears to represent a happy medium compared to other popular stopping criterions, achieving comparable accuracy to the criterions that achieve the highest final accuracies in 77% or fewer epochs, while the criterions which stop sooner do so with an appreciable loss to final accuracy. Additionally, we use this as the basis of a new learning rate scheduler, removing the need to manually choose learning rate schedules and acting as a quasi-line search, achieving superior or comparable empirical performance to existing methods.