Explore a comprehensive lecture on extending the Weisfeiler-Leman graph isomorphism test to Euclidean equivariant machine learning for point clouds. Delve into the development of WeLNet, a novel architecture capable of processing position-velocity pairs and computing functions fully equivariant to permutations and rigid motions. Learn about the theoretical foundations, including the simulation of 2-WL tests via PPGN and the concept of weighted summation. Examine experimental results showcasing WeLNet's state-of-the-art performance on N-Body dynamics and molecular conformation generation tasks. Gain insights into the broader implications for AI in drug discovery and connect with the speaker through the provided resources.
Overview
Syllabus
- Intro + Background
- WL for Euclidean Equivariant ML
- Simulation of 2-WL via PPGN
- PPGN
- Weighted Summation Is All You Need
- WeLNet
- Experiments
- Conclusions
- Q+A
Taught by
Valence Labs