L10 Representation learning and deep generative modelling

Lecture slides (Lect10a Download Lect10a, Lect 10b Download Lect 10b)

Date and venue:  L10a: Feb 14, kl.10:15-12:00, Lecture hall D2 and Feb 17, kl. 8:15-10:00,  Lecture hall K1

                                   L10b: Feb 20, kl. 13:15-15:00,  Lecture hall K1

Session format:
Building on the pre-recorded introduction to deep learning fundamentals (L9) and representation learning (L10a), a more detailed account of selected aspects will be given first and then combined with group discussions and Q&A.
In the session L10b, the scope of representation learning will be extended towards deep generative modelling. 

Before the lecture

Pre-recorded video on representation learning (and lab 4 DBN, if possible)

Please make sure you have watched also pre-recorded material for the previous session L9. 

Discussions 10a Download 10a, 10b Download 10b

After the lecture

Recorded Zoom sessions L9-10a (combined) and L10b

 

Scope, topics covered

  • Data representations
  • Learning representations
  • Generative vs discriminative models
  • Deep Autoencoders, Variational Autoencoders
  • Generative Adversarial Networks

 

Reading material:

  • I. Goodfellow et al.: ch.14, 15, 18 (18.1-18.2), 20 (20.1-20.5)

  • tutorials/survey papers, e.g.:
    • Salakhutdinov, R. (2015) Learning deep generative models. Annual Reviews of Statistics and Its Application, 2, p.361–385.
    • Bengio, Y., Courville, A., & Vincent, P. (2013). Representation learning: A review and new perspectives. IEEE transactions on pattern analysis and machine intelligence, 35 (8), p.1798-1828.
    • Bengio, Y. (2009) Learning deep architectures for AI. Foundations and trends® in Machine Learning, 2.1. p.1-127.