PIML Resources
A starting point for understanding and applying physics-informed machine learning.
What Is PIML?
PIML integrates machine learning with physical laws, governing equations, mathematical models and scientific constraints. The word “physics” refers broadly to the principles governing a system, not only to physics as an academic subject.
Conventional ML and PIML
A conventional model learns mainly from data. A physics-informed model can also be trained to respect conservation laws, differential equations, boundary conditions or other domain constraints.
Core Concepts
- Governing equations
- Initial and boundary conditions
- Physics-based residuals
- Automatic differentiation
- Forward and inverse problems
- Data and physics loss functions
- Uncertainty and model validation
Major Methods
Physics-Informed Neural Networks, neural operators, hybrid physics–data models, differentiable simulation, surrogate models, reduced-order models and scientific foundation models.
Learning Pathway
- Review the mathematical model of the system
- Learn neural-network and optimisation fundamentals
- Understand automatic differentiation
- Implement a simple ODE or PDE example
- Validate the model against analytical or numerical results
- Progress to an engineering research problem
Research Enquiries
Researchers seeking collaborators or guidance in framing a PIML problem may contact the Society.
Advance interdisciplinary research with the PIML community.
Contact us about research collaboration, programs, membership or institutional participation.