IBM PowerAI Training Workshop

February 28, 2018
1040 NCSA building
1205 W. Clark St., Urbana, Illinois

NCSA will soon be installing a powerful system for deep-learning as part of a major research instrumentation (MRI) grant from the National Science Foundation. The purpose of this award is to deploy a multi-node multi-GPU system and develop tools, techniques, and algorithms to better support deep learning applications for the University of Illinois research community. We are evaluating IBM Power8 and Power9 systems for this purpose and can offer access on a limited basis to a Power8 server with four NVIDIA P100 GPUs for those researchers interested in making use and providing feedback to this hardware.

This workshop is designed to help get started with ML and DL on Power8 using PowerAI. Please fill out the registration form if you plan to attend any part of the workshop. Workshop attendance is open to the University of Illinois research community.

Agenda

8:00–8:15 a.m.
Welcome and session objectives

8:15–9:15 a.m.
Lab 1. Introduction to Linear Regression / Introduction to the structure of a Tensorflow program
Hands-on exercise on linear regression: change learning rate, number of epochs, and learning algorithm

9:30–10:30 a.m.
Lab 2. Logistic Regression, Multinomial logistic regression / Softmax regression
Hands-on: Multinomial Logistic Regression using Tensorflow and MNIST

11:00 a.m. – noon
Lab 3. Introduction to Fully Connected Neural Network
Hands-on: Fully Connected Neural Net using Tensorflow

1:00–2:00 p.m.
Lab 4. Recurrent Neural Networks
Hands-on exercise with Time-series data (NLP examples etc.) using Tensorflow

2:30–3:30 p.m.
Lab 5. Introduction to Convolutional Neural Networks and ImageNet
Hands-on: ImageNet exercise using CAFFE

3:30–4:00 p.m.
NCSA review of Illinois Power Deep Learning environment, how to access, etc.