Homework 2

Submission date: December 22th 30th, 2018

Topics

• Backpropagation on arbitrarily deep nets
• Optimization algorithms for deep learning
• Training
• Convolutional networks
• Effect of network architecture on accuracy

The assignment code is available here.

We recommend you use git to clone the repo:

git clone https://github.com/vistalab-technion/cs236605-hw2.git


This will allow you to pull updates from us in the event that they are needed.

Note that there are some updates in the environment.yml file since the previous assignment. From within the hw2 directory, please run

conda env update


To update your conda environment (only new dependencies will be installed).

2018-12-09 [14f9d5c]
In Part 2, the expression for the regularization term of the Vanilla SGD was corrected with an additional factor of $\frac{1}{2}$ to make the notebook consistent with the reference implementation (and the unit test).

FAQ

Make sure to read the getting started page and the guide for using course servers (relevant for part 3).

Q: In Part 2, should the regularization term factor be $\frac{1}{2}\lambda$?
A: Yes, see updates.

Q: Should we comment out the raise NotImplementedError() in the YourCodeNet class in models.py while working on Part 1/2?
A: Yes.

Updated: