Bayesian Framework Embeds Linear Constraints for Machine Learning Predictions

A novel Bayesian approach incorporates linear equality constraints into machine learning, enhancing uncertainty assessments and minimizing constraint breaches. This technique, outlined in the paper 'Learning with Embedded Linear Equality Constraints via Variational Bayesian Inference,' weaves linear relationships between inputs and outputs into the learning framework while fully characterizing predictive uncertainty concerning model parameters and domain expertise. When tested on a single particle battery model that adheres to voltage and energy constraints, this method showed narrower credible intervals and fewer constraint violations than traditional Bayesian neural networks utilizing variational inference. The research can be accessed on arXiv within the computer science and machine learning sections.