Interactive HPC Deep Learning with Jupyter Notebooks

Authors: Wahid Bhimji (Lawrence Berkeley National Laboratory), Steven Farrell (Lawrence Berkeley National Laboratory), Oliver Evans (Lawrence Berkeley National Laboratory), Matthew Henderson (Lawrence Berkeley National Laboratory), Shreyas Cholia (Lawrence Berkeley National Laboratory), Aaron Vose (Cray Inc), Mr Prabhat (Lawrence Berkeley National Laboratory), Rollin Thomas (Lawrence Berkeley National Laboratory), Richard Shane Canon (Lawrence Berkeley National Laboratory)

Abstract: Deep learning researchers are increasingly using Jupyter notebooks to implement interactive, reproducible workflows. Such solutions are typically deployed on small-scale (e.g. single server) computing systems. However, as the sizes and complexities of datasets and associated neural network models increase, distributed systems become important for training and evaluating models in a feasible amount of time. In this poster, we describe our work on Jupyter notebook solutions for distributed training and hyper-parameter optimization of deep neural networks on high-performance computing systems.

Best Poster Finalist (BP): no

Poster: pdf
Poster summary: PDF


Back to Poster Archive Listing