A Systematic Comparison of Hard- and Soft-Constrained Physics-Informed Molecular Machine Learning with the Clapeyron Equation
My rating
This paper provides a systematic comparison of hard-constrained and soft-constrained physics-informed molecular machine learning, where thermodynamic relations are either embedded in the model architecture or added to the training loss. The study targets the integration of fundamental thermodynamic relations, exemplified by the Clapeyron equation, into property prediction tasks for materials. The authors note that although both approaches have been applied to various property prediction tasks, a systematic comparison on experimental data had been lacking.
Tags: