News
Lecture today (July 21): Towards Trust in Neural Action Policies
Written on 21.07.2022 09:42 by Joerg Hoffmann
Hi all,
some information about the lecture today (July 21):
This lecture is not exam relevant. What I aim to do is give you a rough impression of ongoing research in my group: Assuming we have learned a neural network that takes action decisions -- how can we gain trust in these decisions? We are currently pursuing approaches towards explanation, safety verification, failure testing, and visualization. I will briefly cover each of these at an abstract level, elucidate the main ideas, and point to open questions. We are always looking for students in these areas, so this could be your chance for a student assistant job, BSc thesis, MSc thesis (provided you are qualified which involves at least a very good grade in this course, and ideally a very good grade in the AI Planning course in the winter term 22/23).
The lecture will be in zoom only, given the currently quite high infection numbers.
best,
Jörg Hoffmann