Summer Semester 2023, TU München

Organizers: Christian Tomani, Felix Wimbauer, Prof. Dr. Daniel Cremers

E-Mail: bdl-ss23@vision.in.tum.de

The registration is managed via the TUM matching system website help. If you like this course, consider giving it a high priority in the matching system.

To apply for this seminar and get a priority, please also send us an email by February 14 to bdl-ss23@vision.in.tum.de with the title “[Application] <Firstname> <Lastname>”, and attach your CV, transcript, and a filled course application form (rename to "firstname_lastname.xlsx"). Download the template for course application form here: Application template. (Please do not change the file format of the excel template!)

Deep learning models nowadays provide state of the art results and set a new standard for many applications, such as speech recognition, computer vision. As standard supervised deep learning has become more and more established, there has been great interest in more specialised topics.

This course will be focusing on advanced deep learning techniques. The topics will include:

• Uncertainty-aware Machine Learning
• Time Series Models
• Bayesian Deep Learning Models
• Diffusion Models
• Unsupervised Representation Learning (Transformer Pretraining, Object-Centric Learning)
• Generative Adversarial Networks

We will be discussing state-of-the-art research and open issues in the scientific community.

Participants should already have a good understanding of basic machine learning and deep learning concepts and models. Especially, they are required to have taken at least one machine learning related course such as:

• Introduction to Deep Learning
• Machine Learning
• Machine Learning for Computer Vision
• Advanced Deep Learning for Computer Vision / Robotics
• Probabilistic Graphical Models in Computer Vision
• etc.

Participants should be able to take initiatives to plan and maintain a continuous workflow and communicate with tutors efficiently.

As many projects consider theoretical aspects of learning theory, a solid basis as well as interest for mathematics is highly recommended.

Prior experiences with machine learning projects are also a plus.

Note: it is crucial for interested applicants to also send us an e-mail (bdl-ss23@vision.in.tum.de) demonstrating their interest and fulfillment of prerequisites. The details will be explained during the pre-course meeting and available on the course website.

Places will be assigned through the TUM matching system (http://matching.in.tum.de).

This course is directed towards master students and will be held as a block seminar.

• Preliminary meeting: February 2nd at 1pm-2pm, in person (01.10.011, MI Building)
• Kick-Off Meeting: April 20th at 1pm-3pm, in person (01.12.035, MI Building)
• Final Presentations: June 7th from 9am-12noon, 1pm-5pm (02.09.023, MI Building) (can be subject to change)

• Deep Learning, Goodfellow, Bengio, Courville, 2016, http://www.deeplearningbook.org/
• Machine learning: a probabilistic perspective, Murphy, 2012
• The Elements of Statistical Learning, Hastie, Tibshirani, Friedman 2001
• Relevant papers will be announced during the course.