A Method for Fast Single-Network Uncertainty Estimation in Deep Learning Interatomic Potentials

Abstract

Deep learning has emerged as a promising paradigm for predicting molecular and materials properties with high accuracy. A common shortcoming shared by deep learning neural networks, however, is that they only produce point estimates of their predictions and do not come with predictive uncertainties associated with these estimates. Existing uncertainty quantification efforts have primarily leveraged the standard deviation of predictions across an ensemble of many independently trained neural networks. This incurs a large computational overhead in both training and prediction that often results in order-of-magnitude more expensive predictions. Here, we propose a method to estimate predictive uncertainty in deep learning using a single neural network without the need for an ensemble. This allows us to obtain uncertainty estimates with virtually no additional computational overhead over standard training and inference. We demonstrate that the quality of the uncertainty estimates matches those obtained from deep ensembles in the task of atomic force prediction for molecular simulations. We further examine the uncertainty estimates of our methods and deep ensembles across the configuration space of our test system and compare the uncertainties to the potential energy surface of the system. Finally, we study the efficacy of the method in an active learning setting and find the results to match an ensemble-based strategy at order-of-magnitude reduced computational cost.