Argonne scientists will have a strong presence at the Deep Learning on Supercomputers workshop. Co-chaired by Argonne’s Ian Foster, the workshop provides a forum for researchers working at the intersection of deep learning and HPC. Argonne researchers are part of a multi-institutional team that will present “DeepDriveMD: Deep-Learning-Driven Adaptive Molecular Simulations for Protein Folding.” The study provides a quantitative basis by which to understand how coupling deep learning approaches to molecular dynamics simulations can lead to effective performance gains and reduced times-to-solution on supercomputing resources.

A research team from Argonne and the University of Chicago will present “Scaling Distributed Training of Flood-Filling Networks on HPC Infrastructure for Brain Mapping” at the Deep Learning on Supercomputers workshop. The team’s paper details an approach to improve the performance of flood-filling networks, an automated method for segmenting brain data from electron microscopy experiments. Using Argonne’s Theta supercomputer, the researchers implemented a new synchronous and data-parallel distributed training scheme that reduced the amount of time required to train a flood-filling network.