Plans for week 42, October 12-16

Dear all and welcome to FYS-STK. First of all, thx so much for all your excellent work with project 1.  We are now into project 2, and with the weekly exercises from last week on the feed forward part and this week's exercises on implementing the back propagation algorithm, you should have the essential elements for setting up your own neural network code.  This week we will first bring in a reminder during Monday's lecture about the back propagation algorithm and the set up of a neural network code. Thereafter, we will discuss how to use the machinery for regression problems, classification problems and finally how to solve differential equations with neural networks. This is a pretty cool topic.

The flipped classroom session this week is devoted to discuss  how to set up a neural network code and the exercises for this week, namely how to set up the back propagation algorithm and the final updates of the unknown parameters.

The detailed plan is 

Monday's lecture, first part

1. Backpropagation: the chain rule, organised

2. Activation and cost functions: why they come in pairs

3. Universal approximation: what a network can represent

Second part of Monday's lecture

4. Neural networks for differential equations

▶ the equation as the cost

▶ why the activation function matters

▶ boundary conditions

▶ pros and cons vs. standard methods

Tuesday:  flipped classroom and backpropagation in code.

Exercises week 42: deadline Friday October 16.

As always, the material for this week is at https://github.com/EducationalMaterialUiO/MachineLearningUiO/tree/main/doc/WeeklyMaterial/week42

Best wishes to you all,

Amanda, Amund, Brit, Kim, Morten, and Viktor