Week 3

This week we will learn how to build a vanilla dense neural network classifier.

Lecture

Lecture notes are here

Recommended reading material and external resources

Note on backpropagation from CS 231n see here

Details on backpropagation for a simple network are here

Recommended video with many example of computational graphs from CS231n is here

Exercise

The assignment this week contains no programming as this weeks lecture content will be a large part of mandatory assignment 1. Most tasks in the exercise this week is considered background knowledge, but basic derivation skills will be tested at the exam. A supplementary resource that can be useful is linked below (most relevant up until Proposition 10).

 

Assignment (pdf)

Solution (pdf)

Exercise resource (pdf)

Notes about the external resource:

  • Their Jacobian is the transpose of the one presented in the exercise and the lecture.
  • The expression in e.g. eq. (44) is the transpose of the gradient, by their element position convention of the gradient and Jacobian.
Publisert 23. jan. 2020 16:08 - Sist endret 29. jan. 2020 09:21