Physics-Enhanced Machine Learning (PEML, also referred to as hybrid, grey-box modelling or scientific machine learning) focuses on the integration of information that is typically extracted from real-world data, physics-based models and domain and expert knowledge. PEML is a natural evolution of Machine Learning (ML) tackling issues such (i) poor generalization performance and physically inconsistent/implausible predictions; (ii) inability of accounting for and quantifying the different sources/types of uncertainties; (iii) inability of providing explainable and interpretable inferences. PEML strategies constrain the space of admissible solutions and therefore enable robust and interpretable models even with limited data. These approaches can allow for real-time model updating, improved forecasting, and the ability to infer system behavior in the presence of sparse or noisy measurements. PEML strategies are critical for (i) informing the physics describing the underlying dynamical system to be able to analyze, control and predict the wide range of behaviours of the real-world system; (ii) providing fast and accurate solutions of hybrid physics-data models, including governing equations, reduced order models, prediction, forecasting and simulation models (iii) identify experiments that need to be carried out.
We welcome your contributions on advanced techniques and industrial applications showcasing recent progress, strengths and limitations of approaches integrating physics knowledge (first principles, domain knowledge, physics constraints, …) with Machine Learning (ML). Particular interest will be given to contributions focusing on strategies including (but not limited to) those leveraging on observational biases (e.g. data augmentation), inductive biases (e.g. physical constraints), learning biases (e.g. inference/learning algorithm setup), and model form/discrepancy biases (e.g. equation terms describing a partially known physics-based model).
Relevant topics include, but are not limited to, hybrid physics-data strategies (*not just physics models, nor just data-driven only techniques) in which the benefit of combining physics and machine learning is clearly emphasised (if in doubt, please check https://arxiv.org/abs/2405.05987) for
(i) overcoming poor generalisation performance and physically inconsistent or implausible predictions;
(ii) providing explainable and interpretable inferences;
(iii) identifying incorrect data and/or physics biases;
(iv) validating modelling and forecasting;
(iv) quantifying different sources of uncertainty.
Registration is free, but only (TBC) places for in-person attendance are available per day – half of which are reserved to early career researchers. It will be possible to attend virtually without presenting. We are grateful for the support of our sponsors (TBC) which has allowed us to provide sponsored places for in person attendance.
Key Dates:
Environmental Statement Modern Slavery Act Accessibility Disclaimer Terms & Conditions Privacy Policy Code of Conduct About IOP
© 2021 IOP All rights reserved.
The Institute is a charity registered in England and Wales (no. 293851) and Scotland (no. SC040092)