Abstract: TBD. Talk on test-time training and reinforcement learning by Jonas Hübotter (ETH Zürich). Approximately 45 minutes.

How do LLM agents turn plain English into complex action? This talk demystifies their inner loop, tool calls, memory, error handling, and goal pursuit, through a from fundamental deep dive into the mechanics that make real-world agents work. We'll start off which classical RL based agents and use that framework to understand cutting edge current LLM powered agents.

How do LLM agents turn plain English into complex action? This talk demystifies their inner loop, tool calls, memory, error handling, and goal pursuit, through a fundamental deep dive into the mechanics that make real-world agents work. We'll start off with classical RL based agents and use that framework to understand cutting edge current LLM powered agents.

Training models on large-scale data has given us powerful generative capabilities for text, images, and video. However, this success has not yet extended to training generalist embodied agents. This talk tackles this gap by focusing on a potential solution to this problem: scalable world models. We'll trace the idea of planning in predictive models, from its origins to modern efforts on building world models directly from pixels. I'll discuss the primary challenge of scaling these models and present our work, Genie, which enables us to learn world models without explicit action labels at scale, demonstrating a new path forward for training the generalist agents of the future.

Reinforcement learning for games at Supercell

Introduction to reinforcement learning and deep learning concepts.