Teaching‎ > ‎

Neural Networks Summer 17

This is the main website for the Summer 2017 course on Neural Networks. Please come back for announcements.

Lectures: Thursdays 12:15-14:00

Final Exam: Tuesday 20.06.2017 15:00-17:00 in 119.

Assignment grading thresholds (out of 60 assignments + 40 projects): 90: 5.0, 80: 4.5, 70: 4.0, 60: 3.5, 50: 3.0
Final exam grade points (out of 75 + 10% from assignment): 70: 5.0, 62: 4.5, 53: 4.0, 44: 3.5, 35: 3.0

Course rules: course rules.pdf

Ranking list:

Extra office hours

posted Jun 15, 2017, 2:04 PM by Jan Chorowski

Please remember about the final exam for Neural Nets, that will happen on Tuesday, 20.06 at 3pm in 119.

I will host office hours according to the following schedule over the next few days, please come to show your homeworks (my group), discuss projects and ask anything course related on:
Friday, 16.6.2017 2-3pm, room 203
Monday 19.06.2017 12-1pm, room 203
Tuesday 20.06.2017 2-3pm, room 203
Wednseday 21.06.2017, 1-3pm, room 203 - I will grade the exams and you will be able to discuss your grade.

Jan Chorowski 

Lecture 15

posted Jun 8, 2017, 1:03 AM by Jan Chorowski

We will first briefly talk about reinforcement learning: https://github.com/janchorowski/nn_assignments/blob/master/lecture_notebooks/14-rl.ipynb

Then I will discuss the attached review slides.

Lecture 14

posted Jun 8, 2017, 12:57 AM by Jan Chorowski

We have discussed EM and autoencoders.

To read more about EM, please refer to the excellent lecture notes by Andrew Ng: http://cs229.stanford.edu/notes/cs229-notes7b.pdf and http://cs229.stanford.edu/notes/cs229-notes8.pdf

Assignment 6 and end of semester information

posted May 30, 2017, 1:39 AM by Jan Chorowski

Assignment 6 is posted on Github:  https://github.com/janchorowski/nn_assignments/blob/master/assignment6/assignment_6.ipynb. It is worth 6 normal points (for a semester total of 60) and several more bonus points. The due date is Friday 16.6.17 Note that this right after Corpus Christi - a day off. All groups (especially the Thursday one) can submit the solutions up to this date using email and show them later, however please don't abuse this mechanism. You can also submit the solution during Jan Chorowski's office hours (Wednesdays 1-3pm).

The projects are worth 40 points. Please see the project information and course rules for ideas and descriptions. For those willing to complete the full projects and choosing the default  (CIFAR10) project please submit it in two phases:
- a basic version achieving > 75% accuracy and a proposal of a further improvement by 16.06.2017,
- final, improved version by the regular final project deadline (29.06.17).

Jan Chorowski will be away starting 22.06.2017, Adrian Lancucki will handle project submissions for all groups after that date. Final project grades need to be entered into USOS by the end of the examination period (30.06.2017), therefore the last day to hand us the projects will be on Thursday, 29.06.2017, there will be extra office hours posted for porject submission.

The assignments and projects sum up to 100 points. To get a passing grade from the exercises, you must have at least 50 points from project and exercises and a basic CIFAR10 solution. Tentative grade thresholds are: 50: 3.0    60: 3.5    70: 4    80: 4.5    90: 5. The final thresholds will not be higher.

The lecture will end with a final exam that is mandatory for everyone, but I will add 10% of your exercise points towards your exam total. We will choose an examination date during next lecture (1.6.17).

Ideas for projects

posted May 22, 2017, 5:40 AM by Jan Chorowski

For those of you that still have not started to think about the final projects: you should do it now! To help I've compiled a little list of ideas.

Lecture 11

posted May 8, 2017, 6:33 AM by Jan Chorowski   [ updated May 8, 2017, 6:34 AM ]

We made a review of the SVM and then followed two ideas

1. Random, untrained weights can work surprisingly well (and can sometimes be used e.g. for architecture search). Notebooks: Echo State Networks (recurrent network with untrained random weights): https://github.com/janchorowski/nn_assignments/blob/master/lecture_notebooks/09-ESNmackey.ipynb random weights in a two-layer network: https://github.com/janchorowski/nn_assignments/blob/master/lecture_notebooks/11_random_weights.ipynb

Lecture 10

posted May 4, 2017, 2:30 AM by Jan Chorowski

Please use on of the many excellent SVM documents, e.g.:
C Burges A Tutorial on Support Vector Machines for Pattern Recognition: https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/svmtutorial.pdf

Lecture 09

posted May 4, 2017, 2:13 AM by Jan Chorowski

Notebook with RNN demonstrations and LSTM theory is here: https://github.com/janchorowski/nn_assignments/blob/master/lecture_notebooks/09-rnn-demos.ipynb

Assignment 5

posted Apr 20, 2017, 1:56 AM by Jan Chorowski   [ updated May 8, 2017, 5:53 AM ]

Deadline: last classes before 16.05.2017 (yes, it's a Tuesday)
Deadline: last classes before 09.05.2017 (yes, it's a Tuesday)
Please do some of the bonus problems!

1-10 of 23