1 h 5 min
Seminar, Hosted by Perimeter Institute for Theoretical Physics
The Wide and Wonderful World of Optimal Transport Theory in Physics
Kavli Institute for Theoretical Physics
Recording
Abstract
Jessica Howard presents optimal transport as a common language for neural-network ensembles, statistical field theories and exact renormalisation-group flows. Viewing a field theory as a probability distribution over functions makes transport methods useful for numerical nonperturbative calculations and analytical insights. The talk connects these ideas to particle-physics data analysis and to feature learning and training dynamics in asymptotically wide networks.
Topics
- optimal transport
- neural-network ensembles, search World Wide
- statistical field theories, search World Wide
- exact renormalisation-group flows, search World Wide
- probability distribution over functions, search World Wide
- numerical nonperturbative calculations, search World Wide
- particle-physics data analysis, broader topic: data analysis
- feature learning
- training dynamics, broader topic: dynamics
- asymptotically wide networks, broader topic: networks