News
ICS '23 paper on Large-scale Mixture-of-Experts (MoE) Training
Siddharth Singh presented recent work on a hybrid tensor-expert-data parallel framework for training MoEs at ICS 2023. The full paper is available here.
PSSG presents tutorial on Distributed Deep Learning at ISC '23
Abhinav Bhatele and Siddharth Singh will present a tutorial on "Distributed Training of Deep Neural Networks" at ISC 2023 on Sunday (May 21, 2023).
Two paper presentations at IPDPS '23 by PSSG students
Siddharth Singh will present "Exploting Sparsity in Pruned Neural Networks to Optimize Large Model Training" and Josh Davis will present "Porting a Computational Fluid Dynamics Code with AMR to Large-scale GPU Platforms" at IPDPS 2023.
Prof. Bhatele promoted to Associate Professor with tenure
Prof. Bhatele has been granted early tenure and promoted to the rank of Associate Professor, effective July 1, 2022.
Onur Cankur presents profiling tools comparison at ISC '22
Our paper on evaluation of call graph generation by profiling tools will be presented at ISC 2022. The full paper is available here.
Two IPDPS '22 papers to be presented by PSSG students
PSSG's Josh Davis awarded NSF GRFP
Josh Davis, a first year PhD student in our group, has been awarded the prestigious NSF Graduate Research Fellowship.
PSSG's Joy Kitson awarded DOE CSGF
Joy Kitson, a first year PhD student in our group, has been awarded the DOE Computational Science Graduate Fellowship. Read more.
Prof. Bhatele receives NSF CAREER award
Prof. Bhatele has received the NSF CAREER award to develop innovative methods for optimizing the performance of parallel applications and runtimes, and the operational efficiency of supercomputers and HPC clusters.
Cluster '20 Paper on Machine Learning for Tuning MPI Collective Performance
Our paper with Sascha Hunold (TU Wien) and others on using machine learning to predict MPI collective performance was presented at IEEE Cluster in September 2020.
ICS '20 Paper on Modeling GPU Performance
Our work with Jaemin Choi and Sanjay Kale (University of Illinois), and David Richards (LLNL) on modeling the end-to-end performance of GPU codes was presented virtually at ICS 2020. The full paper is available here.