Physics Informed Machine Learning

This event will also be available for virtual attendance via Microsoft Teams. Please use the meeting information below:

Join the meeting now

Meeting ID: 293 960 241 702

Passcode: tYP3La

This workshop continues discussions and explorations started in 2016, 2018, 2020, and 2022, at the four previous editions of the workshop. The achievements of modern Machine Learning (ML), nowadays transitioned into Artificial Intelligence (AI), that relies on automatic differentiation, use of massively parallel architectures, and stochastic programming are uncontested in many common applications that impact our every day life, such as image classification, speech recognition, automated translation, and robotic motion control. However, going hand in hand with these advances, new requirements are emerging as we face more challenging problems in the space of science and engineering. When it comes to scientific applications and discoveries, numerous challenges remain in routine adoption of existing algorithms: lack of interpretability of learned models; limitations to include laws of nature in the learning process; need for very large training data sets; interest in exploring rare but critical regimes where data is scarce or not available at all; and lack of rigorous guarantees, among others.

This workshop seeks perspectives on leveraging the deep connection between ML/AI and physics with the goal to better understand and model static and dynamic physical systems, in two directions. First, we aim at increase in scale, rigor, robustness, and reliability of novel ML/AI methods required for routine use in science and engineering applications. Second, the workshop discussions are directed towards approaches and methods for physical modeling applications where a big-data, black-box approach to ML/AI is only a starting point; instead, we will discuss innovative approaches that extend application-agnostic techniques by incorporating complex constraints imposed by physical principles (e.g., conservation laws, causality, symmetries, entropy principles, etc.). We invite experts in both ML/AI techniques as well as in domain science applications in physics and related disciplines to discuss progress towards this important research direction for LANL and DOE.

Topics

• Statistical learning
• Markov random fields and graphical models
• Deep learning
• Reinforcement learning and stochastic optimal control
• Diffusion models
• Transformer models
• Optimization
• Dynamical systems
• Scientific machine learning
• Applications in physics-rich problems
• Inference and sampling of rare events

Organizing Committee:

Andrey Lokhov (LANL)

Michael Chertkov (University of Arizona)