Explore the intersection of machine learning and classical physics in this 59-minute lecture by Soledad Villar from Johns Hopkins University. Delve into the concept of equivariant machine learning and its applications in optimization under symmetry. Learn about graph neural networks, invariant spectral methods, and community detection. Examine the expressivity of graph neural networks and understand the fundamental theorem of equivariant machine learning. Investigate translation equivalence, general groups, and numerical examples. Gain insights into the structured approach of applying classical physics principles to machine learning algorithms.
Equivariant Machine Learning Structured Like Classical Physics
Overview
Syllabus
Introduction
Motivation
Graph Neural Networks
Recap
Invariant
Spectral Methods
Community Detection
Graph Neural Networks Expressivity
Equivariant Machine Learning
Output
Fundamental Theorem
Translation equivalence
General groups
Numerical examples
Conclusion
Questions
Taught by
Simons Institute
Related Courses
-
Subgraph-Based Networks for Expressive, Efficient, and Domain-Independent Graph Learning
-
Equivariant Machine Learning from Classical Physics
-
Equivariant RL
-
Unsupervised Learning of Group Invariant and Equivariant Representations
-
Understanding Deep Neural Networks - From Generalization to Interpretability
-
Learning Lagrangian Fluid Mechanics with E3 Equivariant Graph Neural Networks
