Day and Date	Time	Topics	Trainers

Day and Date	Time	Topics	Trainers
Wednesday, 29th Jan	1400 - 1530	Introduction to GPU computing	Trainer: Nicolin Support: Yannick/Rafal Open
Wednesday, 29th Jan	1600 - 1730	AI/ML in 90 mins	Trainer: Deepak Slides: Support: Hongyang/Hendrik
Thursday, 30th Jan	0900 - 1030	Hands-on introduction to Model Order Reduction 1	Trainer: Hendrik Support: Deepak
Thursday, 30th Jan	1100 - 1230	Hands-on introduction to Model Order Reduction 2	Trainer: Hendrik Support: Deepak

Topics that will be addressed in the workshops

Introduction to GPU computing

• Introduction: (15 Min)

• Example 1: Making Loops Parallel on GPU (20 Min). ​

• Example 2: Finding Neighbors (30 Min).​

• Advanced Acceleration Methods for Neighbor Finding (10 Min).​

• Open discussion on other implementations to consider @Darius Mačiūnas ( 7 min) .
  (If unable to reach this point in 90 mins, we will continue this in our WG1 meeting/hackathon on Thursday afternoon)

Hardware requirements

This taster does not require you to run any code as you would need a GPU, if you do have a GPU then you are welcome to implement as we go along.

Contributors:

@Yannick Descantes@Rafal Kobylka @Retief Lubbe

Join the meeting now
Meeting ID: 334 188 254 379
Passcode: e3f5Vi2c

AI/ML in 90 mins

• This will be a 90 mins crash course on machine learning and deeeeep learning concepts

• By the end of this session you should be able to know

  • ML terminology/jargon

  • Underlying core ML principle

  • Types of data, learning and algorithms

  • If time permits,

    • On-paper exercise on basic understanding of how a neural network works

    • And, how to setup a DS/ML environment, with special focus on reproducibility

      • If unable to reach this point in 90 mins, we will continue this in our WG1 meeting/hackathon on Thursday afternoon.

Hardware requirements

• Pen and paper

• A laptop with a possibility to connect to the internet.

Hands-on introduction to MOR 1 and 2

Session 1

• We start by looking at parametrised problems, in particular parametrised elliptic PDEs.

• We will learn about the reduced basis method and the efficient construction and evaluation of reduced order models.

• Finally, employ machine learning by using deep neural networks in a data-driven manner will be integrated in order to further speed up the computations.

• We show the implementation of the presented methods within the Python package pyMOR and discuss some of the basic design principles of the software.

Session 2

• Focus is devoted to input-state-output dynamical systems with particular emphasis on linear time-invariant systems.

• We first provide some background information on these systems and discuss their main features.

• We introduce projection-based model order reduction for these systems and consider the well-established balanced truncation method in more detail.

• Again, the application of the pyMOR software package to construct reduced models for control systems is shown.

Hardware requirements

 If you would like to follow along and run the code yourself, you can bring your own laptop with an internet connection and use a binderhub instance (link provided during the training).