Connected: Meifeng, Paolo, Peter, Rui, Walter, Xiangyang
Notes:
- Peter on LBANN
- Application/user POV
- LBANN originally LLNL LDRD
- Start from pyTorch model, hand it over to LBANN team, who does any needed development
- 512**3 density measurements. Each measurement has four channels corresponding to different redshifs (e.g. 500M yrs, 1Byrs, 8By, today)
- MultiGAN is an ensemble of GANs, with each D acting on its G, but with a master G and a master D exchanging weights and detecting and correcting mode collapse.
- Part of LBANN, limited documentation, Brian (main author) would help with your use case.
- Training datasets curated and available on globus
- MultiGAN PyTorch version available, works for small models
There are minutes attached to this event.
Show them.