Berkeley Deep Generative Models for Fundamental Physics Meeting
Wednesday, March 17, 2021
1:00 – 5:00 PM PST
Register to attend this virtual meeting.
Schedule
(Timing links below lead directly to the individual YouTube video presentations.)
- 00:08:00 — Introduction [slides]
- 00:09:10 — 15’+5’ Talk (Colliders), Speaker: Raghav Kansal (UCSD), Title: Graph Generative Adversarial Networks for High Energy Physics Data Generation [slides]
- 00:28:40 —15’+5’ Talk (Nuclear Physics), Speaker: Felix Ringer (LBNL), Title: GANs for parton shower development [slides]
- 00:53:02 —15’+5’ Talk (High Energy Physics), Speaker: Yadong Lu (UC Irvine), Title: Sparse Autoregressive Models for Scalable Generation of Sparse Images in Particle Physics [slides]
- 01:18:05 — 15’+5’ Talk (Cosmology), Speaker: Biwei Dai (UC Berkeley), Title: Normalizing Flows for data with Translational and Rotational Symmetry [slides]
- 01:40:40 — 15’+5’ Talk (Astronomy), Speaker: Jorge Martinez-Palomera (BAERI), Title: Deep Generative Modeling of Periodic Variable Stars Using Physical Parameters [slides]
- 02:01:00 — 15’+5’ Talk (Astrophysics), Speaker: David Shih (Rutgers / LBNL), Title: Via Machinae: Discovering Stellar Streams and Modeling the Galaxy with Normalizing Flows [slides]
- 02:25:20 — 15’+5’ Talk, (CMB), Speaker: Ben Thorne (UC Davis), Title: A Generative Model of Galactic Dust Emission Using Variational Inference [slides]
High fidelity simulations are a foundational component of fundamental physics (particle physics, nuclear physics, astrophysics, cosmology) research. However, these simulations are often too slow or are limited in key ways. As a result, deep generative models have shown great promise for replacing or augmenting various aspects of data analysis in fundamental physics. In the spirit of the likelihood-free inference workshop held about a year ago at the Berkeley Institute for Data Science (BIDS), we are gathering (virtually) at BIDS for an afternoon to exchange problems, ideas, and techniques in the area of deep generative modeling for fundamental physics (co-hosted with the ML group in the Physics Division at Berkeley Lab). Here is an incomplete list of topics we would like to cover:
- Applications of generative modeling to fundamental physics
- Physics-informed models
- Practicalities of training and inference (mode collapse, generation speed, staticial power, …)
- Method comparisons (flows, autoencoders, GANs, …)
- Implicit versus explicit density models (our focus is on generation and not density estimation, but interesting to know if the latter helps with the former)
Propose a Discussion Topic: We aim to make this meeting a collaborative and productive event, and in that spirit we invite participants to propose topics for discussion. Proposed sessions can be found at this link, please comment on the proposals to indicate your interest.
Important dates
Februrary 17, 2021: Opening registration
March 3, 2021: Schedule finalized
March 17, 2021: Workshop
Organizing Committee
Ellianna Abrahms, Department of Astronomy, UC Berkeley; Vanessa Boehm, Department of Physics, UC Berkeley; Aishik Ghosh, UC Irvine / Physics Division, Berkeley Lab; Yue Shi Lai, Nuclear Science Division, Berkeley Lab; Mustafa Mustafa, NERSC, Berkeley Lab; Ben Nachman, Physics Division, Berkeley Lab; Giuseppe Puglisi, Space Science Laboratory, UC Berkeley
Speaker(s)
Benjamin Nachman
Ben Nachman is a Staff Scientist in the Physics Division at LBNL where he is the group leader of the cross-cutting Machine Learning for Fundamental Physics group. He was a Churchill Scholar at Cambridge University and then received his Ph.D. in Physics and Ph.D. minor in Statistics from Stanford University. After graduating, he was a Chamberlain Fellow in the Physics Division at Berkeley Lab. Nachman develops, adapts, and deploys machine learning algorithms to enhance data analysis in high energy physics. He is a member of the ATLAS Collaboration at CERN.