2020-04-16 CANCELLED: Intel® AI HPC Workshop #2: Deep Learning Module (HIDL1S20)

The demand of using Deep Learning techniques in many scientific domains is rapidly emerging and the requirements for large compute and memory resources is increasing. One of the consequences is the need of the high-performance computing capability for processing and inferring the valuable information inherent in the data. The Leibniz Supercomputing Centre (LRZ) has recently installed its new high-end system, SuperMUC-NG. Based on Intel^® technology, it targets among others also workloads at the crossroads of AI and HPC.

In this session you will learn various optimization methods to improve the runtime performance of Deep Learning algorithms on Intel^® architecture. We cover how to accelerate the training of deep neural networks with TensorFlow, thanks to the highly optimized Intel^® Deep Neural Network Library (DNNL). We also demonstrate techniques on how to leverage deep neural network training on multiple nodes on an HPC cluster.

Schedule:

13:30 - 14:30   Deep Learning – Optimized training instances

• Performance Optimized Deep Learning Frameworks solutions from Intel^®
  • TensorFlow and PyTorch optimizations for CPU via Intel^® DNNL
• Distributed (data parallel) deep learning training with Horovod on a CPU cluster
• Large memory (100 GB to 1.5 TB) training with TensorFlow
• Federated Learning with Intel SGX enclaves for security and privacy

14:30 – 15:00 Hands On Session

• Distributed Intel^®-Optimized Tensorflow with Horovod

15:00-15:15    Coffee Break

15:15-16:30    Deep Learning – Optimized inference instances

• Performance Optimized Deep Learning Inference using the Intel^®distribution of the OpenVINO toolkit
  • What is OpenVINO?
  • Case studies from industry
  • Model Serving
  • Creating an inference pipeline for OpenVINO

15:30 - 17:00    Hands On Session

• AI Inference with the Intel Distribution of OpenVINO

Basic knowledge of Python.

Certificates of attendance for

• Introduction to GNU/Linux and SSH (or comparable previous knowledge)
• at least one of Introduction to the LRZ HPC Infrastructure or Introduction to the LRZ Compute Cloud (or comparable previous knowledge)
• Using Python at LRZ (or comparable previous knowledge)

All participants are expected to bring their own laptops.

J. Albert-von der Gönna (LRZ)