Deep Unsupervised Learning - WS24/25

This repo is for THWS/MAI course Deep Unsupervised Learning in WS24/25. The course is largely based on the Berkely course CS294-158-SP24 of Pieter Abbeel and his team. Many thanks to them for releasing all the material.

Instructor: Magda Gregorová

General info:

Background material will be provided through this repo.

Communication primary: https://mattermost.fiw.fhws.de/mai/channels/dul_ws24
(If still not on mattermost: https://mattermost.fiw.fhws.de/signup_user_complete/?id=tca4uuxfeprdxeaeoc57hakqqr&md=link&sbr=su )

Lectures: Wed 10:00-11:30 SHL, Fri 13:30-15:00 CAIRO

Schedule (very tentative)

For each date topics planned for discussion are listed. You should look at the material before to be able to discuss.

• M 9/10
  • Intro slides, video 23:25 - end
  • Autoregressive models: slides, video 00:00 - 26:45
  • HW1 Q1
• F 11/10
  • Autoregressive models: slides, video 26:45 - 1:11:15
  • HW1 Q2 part a
• M 16/10 - rescheduled to Tuesday 15/10 at 16:00 in CAIRO!
  • HW1 Q1 - due
  • Autoregressive models: slides, video 1:11:15 - end
  • HW1 Q3 part a, b, HW1 Q5
• F 18/10 - NO COURSE! - MG away
• W 23/10 - NO COURSE! - day off
• F 25/10
  • HW1 Q2 part a, HW1 Q3 part a, b, HW1 Q5 - Q&A
• W 30/10
  • HW1 Q2 part a - due before the course, discussion in the course
  • HW1 Q3 part a, b, HW1 Q5 - Q&A
  • Latent variable models: slides, video 0:00 - 1:23:25
• F 1/11 - NO COURSE! - day off
• W 6/11
  • HW1 Q3 part a, b, HW1 Q5 - due before the course, discussion in the course
  • HW2 Q1, HW2 Q2 part a - Q&A
• F 8/11
  • HW2 Q1, HW2 Q2 part a - Q&A
  • Latent variable models: slides, video 1:23:25 - end
• W 13/11 - NO COURSE! (test lectures professorship "AI and Cybersecurity" - you are welcome to join)
• F 15/11
  • HW2 Q1 - due before the course, discussion in the course
  • HW2 Q2 part a - Q&A
• W 20/11 - NO COURSE! - take time to move ahead
  • HW2 Q2 part a - due, discussion on request possible via zoom
• F 22/11 - NO COURSE! - faculty away
• W 27/11 - NO COURSE! - take time to move ahead
• F 29/11 - NO COURSE! - take time to move ahead
• W 4/12
  • Generative Adversarial Networks / Implicit Models: slides, video - complete
• F 6/12
  • HW3 Q1, Q2 - Q&A on request - online
• W 11/12
  • HW3 Q1, Q2 - Q&A
• F 13/12 - NO COURSE! - take time to move ahead
• W 18/12
  • Diffusion models: slides, video 0:00 - 51:35
• F 20/12 - NO COURSE! - take time to move ahead
• W 8/1
  • HW3 Q1, Q2 - due and discussion !
• F 10/1 - NO COURSE! - take time to move ahead
• W 15/1
  • HW4 Q1, Q2 - Q&A
• F 17/1 - NO COURSE! - take time to move ahead
• W 22/1
  • HW4 Q1, Q2 - Q&A
• F 24/1
  • ALL HW DUE

Homeworks

HowTo:

1. Fork the repository (see "Working with forks").
2. Work in the forked repository as you like - create branches, commits etc.
3. Once done with the HW, run the HW jupyter notebook from the beginning to the end and commit to freeze the state
4. Sent me via mattermost the link to the specific commit in your fork on GitHub

Evaluation

• all HW shall be submitted by Friday 24/1
• for excellent grade solutions for all obligatory parts shall be solved and the solutions shall be good
• each unsolved exercise (or solved badly) will worsen the grade by one step (from 1 to 1.3, 1.7 etc.)