Cost-effective Reduced-Order Modeling via Bayesian Active Learning

TL;DR

This paper introduces BayPOD-AL, an active learning framework that uses Bayesian uncertainty to efficiently develop reduced-order models, significantly decreasing training data requirements and computational costs in complex system simulations.

Contribution

The paper presents a novel Bayesian active learning method for reduced-order modeling that improves data efficiency and generalizability over existing approaches.

Findings

BayPOD-AL effectively identifies informative data points for training.

It reduces computational costs compared to other active learning strategies.

Demonstrates successful application to temperature evolution prediction in complex systems.

Abstract

Machine Learning surrogates have been developed to accelerate solving systems dynamics of complex processes in different science and engineering applications. To faithfully capture governing systems dynamics, these methods rely on large training datasets, hence restricting their applicability in real-world problems. In this work, we propose BayPOD-AL, an active learning framework based on an uncertainty-aware Bayesian proper orthogonal decomposition (POD) approach, which aims to effectively learn reduced-order models from high-fidelity full-order models representing complex systems. Experimental results on predicting the temperature evolution over a rod demonstrate BayPOD-AL's effectiveness in suggesting the informative data and reducing computational cost related to constructing a training dataset compared to other uncertainty-guided active learning strategies. Furthermore, we demonstrate BayPOD-AL's generalizability and efficiency by evaluating its performance on a dataset of higher temporal resolution than the training dataset.