Foundation models and generative AI

Next, we will discuss how foundation models can be adapted beyond in-context learning. Building on the concepts introduced so far, we will explore strategies for adaptation, fine-tuning, and test-time training. These approaches allow models to specialize to new domains or tasks, improve efficiency, and remain robust under distribution shifts.

Part of this week's lecture will be a guest lecture by Jonas Hübotter and Andreas Krause from ETH Zurich. Lecture and guest lecture take place as usual in PO 01.

Important! Assignment 2 is released. Please find below the instructions and information required for the submission. Assignment 2 is due December 3, 23:59.