Course syllabus

Jump to today

Welcome to the course pages for EDAP30 / EDAF90F - Advanced Applied Machine Learning

Links to the official course plan in Swedish and English.

Note that information on this page might change somewhat, especially re exact dates. Anything that looks weird might be due to a miss when transferring information from last year's course occasion - please ask in such a case!

General information

The course will start on Monday, March 18th, all places will be filled according to the rules regarding prerequisites and ranking as explained below.

The course aims to give a deeper insight into ANN-based (and related) techniques for text and image processing (RNNs and CNNs and beyond), as well as to introduce Reinforcement Learning and Bayesian Optimisation.

Registration and admission information

2024-03-10: To keep your seat in the course (if notified that you have one), you need to 

  • Register by March 14th
  • Attend the first lecture on March 18th (or notify Elin (elin_a.topp@cs.lth.se) or Erik (erik.hellsten@cs.lth.se) that you cannot make it, note that you should have a valid reason for not attending like illness, a work schedule you could not modify or study-related travels).
  • Sign up for and attend the first lab session on Thursday, March 21st or Friday 22nd.
  • Join the Discord channel: https://discord.gg/TZpFyDzAu5.

 

Registration information for Ph.D. students (EDAN95F):

2024-03-10: Contact Elin and Erik by e-mail (elin_a.topp@cs.lth.se or erik.hellsten@cs.lth.se) and we will let you know what the situation is re places.

 

Communication during the course:

As for EDAN96, we will use Discord as a means of communication and to organize help / presentation sign-up in lab sessions. You find the Discord link on the page Assignments and lab sessions, general info (Page with lab session info)

Syllabus

Sessions and assignments: The course has 14 lectures, 7 lab sessions (some are compulsory), 4 programming assignments (labs) that should be worked on in pairs, and 3 individual homework assignments (reports) that correspond to three of the programming assignments. To keep your seat in the course, you need to attend the first lab session (technical setup support). See Assignments and lab sessions, general info (page with info on lab sessions) to learn more about the lab sessions.

Modules: The course has 5 Canvas modules (0-4), grouping related lectures and assignments.
Module 0 (EH) gathers the first and last lectures and gives general information about the course.
Modules 1-4 contain 3 lectures each, and have the course content distribution of the following topics and methods to be discussed:

  • Module 1 (EH): Gaussian processes, Bayesian optimization, Hyperparameter optimization, etc.
  • Module 2 (PN): Text processing and RNNs, LSTMs, Transformers, etc;
  • Module 3 (MK): Image classification and CNNs, etc;
  • Module 4 (VK): Action learning and Markov decision processes, RL, etc;

Lectures will be held by four teachers, Erik Hellsten (EH, course responsible), Pierre Nugues (PN), Marcus Klang (MK), and Volker Krüger (VK).

Teaching assistants: For the lab sessions and to support you in general with the course material, we have four TAs, all are Ph.D. students: Faseeh Ahamd, Carl Hvarfner, Konstantin Malysh, and Leonard Papenmeier.

Contact information and office hours for teachers and TAs can be found on the Contacts page (Contacts).

Examination: Fulfilling all assignments (programming assignments in pairs, homework assignments individually) will give the full course credits (7.5 ECTS) and a passing grade "3", as well as allow you to take the optional exam, with which it is then possible to get a higher grade ("4", or "5"). Fulfilling respectively marked (more complex) parts of assignments can generate bonus points for the exam.

If you are a Ph.D. student registered for EDA090F, fulfilling all assignments including potential extra parts (programming assignments in small groups including discussing them in a seminar session (upon agreement), homework assignments individually) will give the full course credits (7.5 ECTS) and a passing grade (G for "godkänd / passed") on the U/G scale. 


Overview of the course events per course/calendar week (will be updated continuously):

week / cw Lectures (topics)

Programming assignments (Labs) due

 Assignment announced Homework due
1 / 12 18/03: Introduction to the course, some recap of EDAN96 (EH) (Module 0)
20/03: Model selection (EH) (Module 1)

22/03: Tech setup (OBS, compulsory)

 Assignment 1 (Lab & report)
2 / 13 25/03: Kernels and Gaussian processes (EH) (Module 1)
27/03: Bayesian Optimization (EH) (Module 1)

 
Spring break and Exam week  

 

  
4 / 16 15/04: Language models and embeddings (PN) (Module 2)
17/04: Recurrent networks (PN) (Module 2)

19/4: BO programming Assignment

 Assignment 2 (Lab & report)
5 / 17

22/04: Transformers (PN) (Module 2)
24/04: Image classification with CNNs (MK) (Module 3)

Assignment 3 (Lab & report)

 

 26/4: Report 1
6 / 18

29/04: CNN Architectures (and transfer learning) (MK) (Module 3)
02/05: Training CNNs (+augmentation) (MK) (Module 3)

03/05: Sequential Networks
7 / 19

06/05: Markov decision processes and Reinforcement Learning (VK) (Module 4)
08/05: Reinforcement Learning (VK) (Module 4)

10/05: Image Classification with CNNs

 Assignment 4 (Lab & report)

10/05: Report 2
8 / 20 13/05: TD Learning (VK) (Module 4)
15/05: Wrap-up (EH) (Module 0)		 17/05: Reinfocement Learning 17/5: Report 3

 

Course summary:

Date Details Due