Hybridizing Physical and Data-driven Prediction Methods for Physicochemical Properties

TL;DR

This paper introduces a hybrid approach combining physical models and data-driven methods via Bayesian inference to improve predictions of physicochemical properties, demonstrating significant accuracy gains over existing methods.

Contribution

The paper presents a novel hybridization technique that distills physical model predictions into a prior and integrates sparse data, enhancing prediction accuracy.

Findings

Significant improvement over physical and data-driven baselines

Effective hybridization of physical and data-driven models

Enhanced prediction of activity coefficients at infinite dilution

Abstract

We present a generic way to hybridize physical and data-driven methods for predicting physicochemical properties. The approach `distills' the physical method's predictions into a prior model and combines it with sparse experimental data using Bayesian inference. We apply the new approach to predict activity coefficients at infinite dilution and obtain significant improvements compared to the data-driven and physical baselines and established ensemble methods from the machine learning literature.