The Argonne Leadership Computing Facility will host a hands-on training session on DeepHyper (https://github.com/deephyper/deephyper), a distributed automated machine learning (AutoML) software package for automating the design and development of deep neural networks for scientific and engineering applications.
DeepHyper seeks to bring a scientifically rigorous automated approach to neural network model development by reducing the manually intensive, cumbersome, ad hoc, trial-and-error efforts.
DeepHyper can run on a laptop, medium range clusters, and supercomputers with thousands of compute units (GPUs).
The workshop is organized by:
- Prasanna Balaprakash, R&D Lead and Computer Scientist, MCS Division and ALCF, Argonne National Laboratory
- Romain Égelé, Graduate Student and Research Aide, Université Paris-Saclay and Argonne National Laboratory
- Romit Maulik, Assistant Computational Scientist, MCS Division, Argonne National Laboratory
- Bethany Lusch, Assistant Computer Scientist, ALCF, Argonne National Laboratory
- Krishnan Raghavan, Assistant Computational Mathematician, MCS Division, Argonne National Laboratory
- Anirudh Subramanyam, Postdoctoral Researcher, MCS Division, Argonne National Laboratory
- Sandeep Madireddy, Assistant Computer Scientist, MCS Division, Argonne National Laboratory
- Tanwi Mallick, Assistant Computer Science Specialist, MCS Division, Argonne National Laboratory
- Akhil Pandey Akella, Graduate Research Assistant, Northern Illinois University
- Nesar Soorve Ramachandra, Computational Scientist, CPS Division, Argonne National Laboratory
- Kyle Felker, Assistant Computational Scientist, CPS Division, Argonne National Laboratory
- Sam Foreman, Assistant Computational Scientist, ALCF, Argonne National Laboratory
EVENT DATE
The virtual training is scheduled Friday July 15, 2022 from 9:00 a.m. - 4:00 p.m. US Central.
Registration Deadline: Friday, July 8, 2022
EVENT DETAILS
In this virtual workshop attendees will learn various capabilities of the DeepHyper software to automate the design and development of neural networks.
DeepHyper GitHub repo: https://github.com/deephyper/deephyper
DeepHyper Documentation: https://deephyper.readthedocs.io/en/latest/
8:30 -- 9:00: Zoom Check-in and setup
9:00 -- 9:10: Welcome & Intro (Prasanna Balaprakash)
9:10 -- 9:30: Hyperparameter search (Prasanna Balaprakash)
9:30 -- 10:00: Hands-on (Romain Égelé)
10:00 -- 10:20: Neural architecture search (Romit Maulik)
10:20 -- 10:50: Hands-on (Romit Maulik)
Break for 10 mins
11:00 -- 11:20: Ensembles & uncertainty quantification (Bethany Lusch and Krishnan Raghavan)
11:20 -- 11:50: Hands-on (Bethany Lusch and Krishnan Raghavan)
Lunch Break and Q & A Session for 40 mins
12:30 -- 12:50: Multiobjective hyperparameter search (Anirudh Subramanyam)
12:50 -- 1:20: Hands-on (Anirudh Subramanyam)
1:20 to 1:40: Transfer learning in hyperparameter search (Sandeep Madireddy and Tanwi Mallick)
1:40 -- 2:10: Hands-on (Sandeep Madireddy and Tanwi Mallick)
Break for 10 mins
2:20 to 2:40: Graph neural architecture search for molecular property prediction (Akhil Pandey Akella)
2:40 -- 3:10: Hands-on (Akhil Pandey Akella)
3:10 -- 4:00: Running DeepHyper at Scale on Perlmutter/NERSC (Nesar Soorve Ramachandra), ThetaGPU/ALCF (Kyle Felker/Sam Foreman), Summit/OLCF platforms (Kyle Felker) No hands-on; only demo.
All hands-on will be on Google collab.
Participants are expected to have basic working knowledge on neural network development using Tensorflow, Keras, and/or PyTorch.
