Machine Learning on HPC Systems is workshop series held in conjunction with ISC Supercomputing Conference
- 2020: papers and videos of the talks given at MLHPCS'20 can be found here
- the second workshop on Machine Learning on HPC Systems is an online event, held in conjunction with the ISC 2021 (July 2nd).
Over the last few years, Machine Learning (ML) - and in particular Deep Learning (DL) - has become an important research topic in the High Performance Computing (HPC) community. This comes along with new users and data intensive applications on HPC systems, which increasingly affects the design and operation of compute infrastructures. HPC environment and resources on the one hand provide opportunities to attack ML/DL problems not tractable otherwise. On the other hand, the ML/DL community is just getting started to utilize the performance of HPC, leaving out many opportunities for better parallelization and scalability. The intent of this workshop is to bring together researchers and practitioners from all communities to discuss three key topics in the context of High Performance Computing and ML/DL methods: parallelization and scaling of ML / DL algorithms, ML/DL applications on HPC systems, and HPC systems design and optimization for ML / DL workloads.
Date: in conjunction with the ISC 2021 online conference
- ISC'21 is held from June 24th until July 2nd
- Workshop date: July 2nd
The aim of the workshop is to provide a platform for technical discussions and the presentation of work in progress, as well as, unsolved problems, which is complementary to the “Machine Learning Day” in the main conference program.
- Unsolved problems in ML / DL on HPC systems
- Scalable ML / DL algorithms
- Parallelization techniques
- Libraries for ML / DL
- Tools + workflows for ML / DL on HPC systems
- Optimized HPC systems design and setup for efficient ML / DL
- ML / DL applications on HPC Systems
Program Fr, July 2nd 2pm-6pm MESZ (local time Frankfurt, Germany)
2pm Welcome and Intro
2.15pm Keynote: Surprises in Deep Learning Training: Symmetry, Efficiency, and Scale
Daniel Soudry, Technion [abstract]
3pm Invited Talk: Impact of large-scale pre-training on intra- and inter-domain transfer learning in full and few-shot regimes
Jenia Jitsev, Juelich Supercomputer Center, Helmholtz AI, Research Center Juelich [abstract]
3.30pm Invited Talk: Large-scale Neural Solvers for Partial Differential Equations
Nico Hoffmann, TU Dresden [abstract]
4pm Contributed Talk: MSM: Multi-Stage Multicuts forScalable Image Clustering
Kalun Ho (Fraunhofer ITWM); Avraam chatzimichailidis (Fraunhofer ITWM ); Margret Keuper (University of Mannheim); Janis Keuper (Fraunhofer ITWM + IMLA, Offenburg University) [abstract]
4:15pm Contributed Talk: Analysis of Black-box Optimization Algorithms to Tune TensorFlow's CPU Backend
Niranjan Hasabnis (Intel); Derssie Mebratu (Intel)[abstract]
4.30pm Invited Talk: Deep Learning Meets Optimal Control - How Optimal Control Enables Faster and Better Training
Stefanie Günther, LLNL [abstract]
5pm Invited Talk: Challenges when scalling DL training to thousands on GPUs and TPUs
Ahmed Elnaggar, TU Munich [abstract]
5.30pm Contributed Talk: Parallel/distributed intelligent hyperparameters search for generative artificial neural networks
Mathias Esteban (Universidad de la República); Jamal Toutouh (Universidad de Málaga); Sergio Nesmachnow (Universidad de la República) [abstract]
5.45pm Contributed Talk: Machine learning for generic energy models of high performance computing
Jonathan Muraña (Universidad de la República); Carmen Navarrete (Leibniz Supercomputing Center); Sergio Nesmachnow (Universidad de la República) [abstract]
6pm Contributed Talk: Hyper-parameter optimisation on HPC – a comparative study
Peter Winkler (TU Dresden); Norman Koch (TU Dresden) [abstract]
6:15pm Closing Discussions
- Juan J. Durillo (LRZ, Munich)
- Dennis Hoppe (HLRS, Stuttgart)
- Jenia Jitsev (JSC, Jülich)
- Janis Keuper (IMLA / Fraunhofer ITWM)
- Sunna Torge (ZIH, Dresden)