Doctoral course - Self-Supervised Learning for Multimodal Data: From Models to Loss Functions

The course gives an introduction to self-supervised learning methods for learning representations of single- and multiple-modality data, covering deep architectures, training target and loss functions used in state-of-the-art methods, and selected downstream applications. A focus will be given to loss functions including both contrastive and predictive losses. The course will be taught through a combination of lectures and mini-projects.

Event information

Time

-

Location

TS127

Add event to calendar

External teacher(s):     Zheng-Hua Tan 
External teacher(s) organization:   Aalborg University
ECTS -credits: 2
Grade: Pass or Fail
Assessment: oral exam after all lectures and research report within four weeks after lectures.

Lecture hall TS127    
(Estimated) Date of the course  *May 15-17, 2023
May 15, 9-11, Monday afternoon: Lecture 1
May 16, 9-11, Tuesday morning: Lecture 2
May 16, 13-15, Tuesday afternoon: Lecture 3
May 17, 9-11, Wed. morning: Lecture 4

May 17, 13-16, Wed. afternoon: Oral exam
         
Learning objectives and contents: 
The course gives an introduction to self-supervised learning methods for learning representations of single- and multiple-modality data, covering deep architectures, training target and loss functions used in state-of-the-art methods, and selected downstream applications. A focus will be given to loss functions including both contrastive and predictive losses. The course will be taught through a combination of lectures and mini-projects.
 
Tentative topics:
Self-supervised learning concept and history
Deep learning architectures 
Key methods: training target and loss functions
Applications
 
Amount of contact teaching hours  *8 hours lecture

Last updated: 12.1.2023