Symmetries, flat minima, and the conserved quantities of gradient flow
- Bo Zhao
- Iordan Ganev
- et al.
- 2023
- ICLR 2023
Bio
Nima Dehmamy is a Research Staff Member at IBM Research, Cambridge MA. His research focuses on AI for science, equivariant networks, optimization theory and graph neural networks. His recent projects include
- Symmetries, conserved quantities and flat minima in the loss landscape
- Generative adversarial symmetry discovery
- Coarse-graining in machine learning models
He earned his PhD in physics from Boston University in 2016. He was a research faculty at Northwestern University 2019-2022.
Publications
Accelerating Network Layouts Using Graph Neural Networks
- Csaba Both
- Nima Dehmamy
- et al.
- 2023
- Nature Communications
Symmetry Teleportation for Accelerated Optimization
- Bo Zhao
- Nima Dehmamy
- et al.
- 2022
- NeurIPS 2022