Seminar: 2.01.812 Theoretical Foundations of AI Safety - Details

Seminar: 2.01.812 Theoretical Foundations of AI Safety - Details

General information

Course name Seminar: 2.01.812 Theoretical Foundations of AI Safety
Subtitle inf812
Course number 2.01.812
Semester SoSe2026
Current number of participants 1
expected number of participants 12
Home institute Department of Computing Science
Courses type Seminar in category Teaching
First date Wednesday, 08.04.26, 10:00 - 12:00 o'clock A03 2-209
Type/Form S
Lehrsprache englisch

Rooms and times

A03 2-209

  • Wednesday, 10:00 - 12:00, Weekly (from 08.04.26)

Module assignments

Comment/Description

This master-level seminar explores the theoretical principles underlying the safety, robustness, and alignment of modern AI systems. Students will engage with cutting-edge research on adversarial robustness, distribution shift, formal verification of neural networks, safe reinforcement learning, and foundational alignment theory.

Each week focuses on a topic, with students leading presentations and discussions to critically analyze assumptions, guarantees, and limitations of current approaches. The seminar emphasizes rigorous reasoning, mathematical foundations, and formal models of safety-relevant behavior. Students will gain a deep understanding of the theoretical challenges in making learning-enabled autonomous systems reliable, predictable, and certifiably safe.

We expect the students to deliver a 1.5~2-hour lecture on a specific topic, making minimal assumptions about participants' knowledge of linear algebra, probability, and statistics. After the lecture, the participants can fully understand the underlying principles while using examples serving as intuitions.

This course is suitable for those who wish to enter the frontier of artificial intelligence research or join leading industrial labs. The solid mathematics and analytical skills will be a great plus for your future career.