EvoSim

PI Name Walter Hopkins
PI Institution Argonne National Laboratory
Project Description

We are attempting to use a genetic algorithm to optimize deep neural networks to generate simulations of the ATLAS (at the LHC) detector response. This hyperparameter search will require the training of many neural networks which is compute intensive and would greatly benefit from the use of GPUs. This project is in preparation for more complex models that will make use of A21 accelerators for training.

Testbed

gpu_v100_smx2 and DGX2