Challenges and Opportunities for Integrating Physics-Knowledge in Machine Learning Strategies

Explore the challenges and opportunities of integrating physics knowledge into machine learning strategies in this comprehensive lecture. Delve into four different biases that can be enforced in ML algorithms: observational, inductive, learning, and discrepancy. Examine a case study on identifying friction forces in dynamical systems, showcasing three physics-enhanced ML strategies based on sparse system identification and Gaussian Processes. Analyze numerical and experimental results demonstrating the applicability of these approaches. Gain insights from Dr. Alice Cicirello, an experienced researcher in dynamics, vibration, and uncertainty, as she discusses open challenges and future directions in this field. Learn how physics-enhanced ML can improve generalization in small-medium data regimes and ensure physics-consistent predictions in large data scenarios.

DDPS | Challenges and opportunities for integrating physics-knowledge in machine learning strategies