Physics- and data-driven Active Learning of neural network representations for free energy functions of materials from statistical mechanics
This work addresses the problem of efficient and accurate free energy representation for materials scientists, though it appears incremental as it builds on existing Active Learning and neural network methods.
The researchers tackled the challenge of accurately modeling free energy functions for materials by developing an Active Learning framework that combines physics- and data-driven approaches to optimize neural network training on DFT-informed Monte Carlo data, achieving reduced mean squared error (MSE) while minimizing required data points.
Accurate free energy representations are crucial for understanding phase dynamics in materials. We employ a scale-bridging approach to incorporate atomistic information into our free energy model by training a neural network on DFT-informed Monte Carlo data. To optimize sampling in the high-dimensional Monte Carlo space, we present an Active Learning framework that integrates space-filling sampling, uncertainty-based sampling, and physics-informed sampling. Additionally, our approach includes methods such as hyperparameter tuning, dynamic sampling, and novelty enforcement. These strategies can be combined to reduce MSE,either globally or in targeted regions of interest,while minimizing the number of required data points. The framework introduced here is broadly applicable to Monte Carlo sampling of a range of materials systems.