Robust Hybrid Learning With Expert Augmentation

TL;DR

This paper introduces expert augmentation, a hybrid data augmentation strategy that enhances the generalization of hybrid models by leveraging expert knowledge, validated through experiments on dynamical systems and a real double pendulum dataset.

Contribution

It proposes a novel expert augmentation method for hybrid models, improving their robustness and generalization beyond the training distribution.

Findings

Expert augmentation improves model generalization in dynamical systems.

Validated on ordinary and partial differential equations.

Effective on real-world double pendulum data.

Abstract

Hybrid modelling reduces the misspecification of expert models by combining them with machine learning (ML) components learned from data. Similarly to many ML algorithms, hybrid model performance guarantees are limited to the training distribution. Leveraging the insight that the expert model is usually valid even outside the training domain, we overcome this limitation by introducing a hybrid data augmentation strategy termed \textit{expert augmentation}. Based on a probabilistic formalization of hybrid modelling, we demonstrate that expert augmentation, which can be incorporated into existing hybrid systems, improves generalization. We empirically validate the expert augmentation on three controlled experiments modelling dynamical systems with ordinary and partial differential equations. Finally, we assess the potential real-world applicability of expert augmentation on a dataset of a real double pendulum.