inf5400 - Advanced Topics in Applied Deep Learning (Complete module description)

This module is intended for an advanced audience and requires a solid understanding of the fundamentals of Machine Learning. Experience in training deep neural networks is essential in this context.

Professional competence
The students

• have in-depth knowledge of selected application areas of deep learning. They are familiar with various solutions for problems in these areas, know their advantages and disadvantages, and can practically implement them and adapt them to their own issues.

Methodological competence
The students

• independently develop theoretical and practical concepts with the help of in-person events, provided materials, and specialized literature.

Social competence
The students

• can present solution approaches for problems in this area to the plenary and defend them in discussions.

Self-competence
The students

• are able to assess their own subject-specific and methodological competence. They take responsibility for their competence development and learning progress and reflect on these independently. In addition, they independently work on learning content and can critically reflect on the content.

General competence goals
++ knowledge of data science/ML methods and its foundations
++ ability to analyze problems, compare and select solution methods
++ formalizing problems mathematically, developing and implementing solutions, interpret
their results
+ data management and infrastructure skills
+ data presentation & discussion
+ Scientific literature (reading & writing)
+ scientific communication skills (in particular with people outside the field of study)

The thematic areas to be covered in various instantiations of the module include deep learning
methods for time series analysis and deep unsupervised learning.
This instatiation on deep generative models will two primary subjects: self-supervised learning
and modern generative models. In the first part, we will examine the fundamental design
principles (contrastive versus non-contrastive) underlying self-supervised learning algorithms.
In the second part, we will explore applications of these principles to specific data modalities
such as computer vision, natural language processing (including an extensive coverage of
large language models) and audio/time series. Finally, the third part will focus on generative
models, where we will cover a wide array of models, ranging from autoregressive models,
variational autoencoders, and normalizing flows, to generative adversarial networks and
(latent) diffusion models.

> Prince, S. J. (2023). Understanding deep learning. MIT press.
> Dawid, A. & LeCun, Y. (2023). Introduction to Latent Variable Energy-Based Models: A Path Towards Autonomous Machine Intelligence. Les Houches Summer School on Statistical Physics and Machine Learning in 2022 https://arxiv.org/abs/2306.02572

At the end of the lecture period

Written exam / oral exam / project work

Active participation: handing in exercises.