Upscaling A.I. with Containers

Overview

Active practitioners in deep learning primarily develop and train their networks on local computing devices. Nonetheless, the accuracy of such networks is dependent on their depth and the amount of training data. Moreover, since local computing devices such as laptops or clusters provide limited computational capacity, training a deep neural network (NN) requires access to large clusters, i.e., supercomputers.

In this workshop, we will learn the basics of containers, how to work with Docker and Singularity, and how to run distributed training using the Horovod framework on a supercomputer.

The workshop will be entirely online using zoom.

Outcomes

• You will be able to create, deploy, and update containers locally or on a cluster.
• You will train, and upscale traditional Convolutional NN (CNN) in TensorFlow.
• You will learn to upscale CNN using Horovod.

Prerequisites

Basic knowledge of UNIX OS and familiarity with NNs are required.

Preliminary Agenda

Day 1 – Tuesday 1 February 2022

Day 2 – Wednesday 2 February 2022

Day 3 – Thursday 3 February 2022

Registration

To register for this event, please follow this link https://events.prace-ri.eu/event/1292/registrations/952/.

For questions regarding this event please contact us at training@enccs.se.

————
This training is intended for users established in the European Union or a country associated to Horizon 2020.