Plans for week 40, September 28-October 2

Dear all, welcome back to FYS-STK. We hope you had a lovely weekend. 

The plans this week are:

Monday lecture:

▶ First 15 min, the reminder from last week: logistic regression drawn as a single neuron and its learning rule (Section 5.8.1); the same picture read backwards the computational graph, forward and reverse sweeps, and why every framework runs reverse mode (Section 5.8.2)

▶ Neural networks and deep learning: the artificial neuron, why one layer is not enough (XOR), multilayer perceptrons, the universal approximation theorem and what it does not say, activation functions, and the cost we have to minimize (Sections 8.18.7)

▶ Back propagation from the smallest networks: one neuron with a scalar input, then one hidden neuron in front of it, by hand and with jax.grad; our rst training program; then two inputs and two hidden nodes, derivative by derivative (Sections 8.8.18.8.2)

▶ Next week: the general case, the four equations of back propagation and the algorithm, and how to write your own network code (Sections 8.8.38.8.5 and 8.13)

Tuesday (fipped classroom): your questions, then only Project 1 two cases on the analysis and presentation of your results. We continue with these discussions during the lab session on Wednesday. No weekly exercises this week: the project is due Monday October 5.

Reading: Chapter 8, Sections 8.18.8; Section 5.8 once more. Goodfellow et al. Chapter 6; Raschka et al chapter 11

The material is at https://github.com/EducationalMaterialUiO/MachineLearningUiO/tree/main/doc/WeeklyMaterial/week40.

Best wishes with the finalization of project 1.

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