PRACE Winter School 2019 - Introduction to Machine Learning for scientists

Mar 4, 2019, 1:00 PM → Mar 8, 2019, 1:00 PM Europe/Brussels

Opleidingcentrum C (University of Leuven - ICTS)

Jan Ooghe (KU Leuven), Geert Jan Bex (University of Hasselt)

Machine learning is increasingly being used as a tool to explore, understand, and analyze the data that is generated in a wide variety of scientific domains. Classic approaches to data processing  start to break under the avalanche of data produced by modern measurement and diagnostic techniques. In the first part of this school researchers will get an overview and a solid understanding of machine learning algorithms. This will show these methods can be used to solve various research questions and remain on the cutting edge of the research domain. The second part will build on this knowledge, and bring applications to the next level by introducing the technology to perform machine learning analysis at scale.

The PRACE Winter School targets at (40) industrial users, researchers and students.

During the hand on sessions attendees will be using the KU Leuven HPC infrastructure (Genius), which is part of the Flemish Computer Center (VSC). The Genius cluster has a total 116 nodes with 2 Intel Gold 6140 (Skylake) processors  and 192 GB of memory. The hands-on will use the GPU section of this cluster, consisting of 20 nodes with 4 NVidia Pascal P100 each.

At the software level students will use a separate conda environment. The advantage of using this kind of environment is that provides trainees with the same software stack both on their laptop and on the HPC cluster.

The program is free of charge (not including travel and accommodation). For the hands-on sessions, participants are expected to bring their own laptops.

Applications are open to researchers, academics and industrial researchers residing in PRACE member countries, and European Union Member States and Associated Countries. All lectures and training sessions will be in English.

Monday, March 4

Introduction to AI and applicability

How does deep learning work: From linear regression to neural networks

3:00 PM Coffe break

Hands-on: Linear regression and fully connected NN

Tuesday, March 5

Supervised learning: classification: Convolutional Neural Networks

Supervised learning: classification: Hands-on: CNN

10:30 AM Coffe break

Supervised learning: classification: Hands-on: CNN

Supervised learning: classification: Frameworks and Software ecosystem

12:00 PM Lunch

Supervised learning: classification: Frameworks and Software ecosystem: Hands-on: Frameworks

Supervised learning: classification. Recurrent Neural Networks

3:00 PM Coffe break

Supervised learning: classification. Recurrent Neural Networks: Hands-on: RNN

Supervised learning: classification. Alternative NNs

Social event: Guided walking tour in Leuven

Wednesday, March 6

From Training to inference. The life of a model: Looking at Inference. Loading Models

From Training to inference. The life of a model: Hands-on and Demo

10:30 AM Coffe break

Unsupervised learning: clustering: Methods and algorithms

12:00 PM Lunch

Unsupervised learning: clustering: Hands-on

Unsupervised learning: dimensionality reduction: Methods and algorithms

3:00 PM Coffe break

Unsupervised learning: dimensionality reduction: Hands-on

Thursday, March 7

Reinforcement learning: How does it work and when to use it

10:30 AM Coffe break

Reinforcement learning: Hands-on

12:00 PM Lunch

Distributed training: Method proposal

Distributed training: Hands-on

3:00 PM Coffe break

AI Ethics and possible future scenarios

Friday, March 8

Deep learning using multiple nodes with GPUs: Emphasis on efficiency and scaling

Deep learning using multiple nodes with GPUs: Hands-on

10:30 AM Coffe break

Deep learning diagnostics/debugging

Wrap up