Test-Time Training Agents to Solve Challenging Problems

We’re thrilled to welcome Jonas Huebotter from ETH Zurich to our joint heidelberg.ai / NCT Data Science Seminar on November 5th at 6 PM.

"When solving a problem of interest, do not solve a more general problem as an intermediate step. Try to get the answer that you really need but not a more general one." — Vladimir Vapnik

In this online session, Jonas Huebotter will guide us through the world of test-time training — a setting in which the model is adapted for each new input during the prediction phase. This unlocks powerful new applications by letting the model adapt and search for the most relevant information to solve the task at hand for this specific input. This is one of the most promising strategies on the ARC-AGI challenge.

We look forward to your participation in this important discussion on the future of trustworthy, privacy-conscious AI that can truly assist us in our private and professional lives.

Abstract

The standard paradigm of machine learning separates training and testing. Training aims to learn a model by extracting general rules from data, and testing applies this model to new, unseen data. We study an alternative paradigm where the model is trained at test-time specifically for the given task. We investigate why such test-time training can effectively specialize a model to individual tasks. Further, we demonstrate that such test-time training enables models to continually improve and eventually solve challenging tasks, which are out of reach for the initial model.

Biography

Jonas Huebotter is a PhD student in the Learning and Adaptive Systems Group at ETH Zurich working with Andreas Krause. Prior to this, he obtained a Master’s degree in Theoretical Computer Science and Machine Learning from ETH Zurich and a Bachelor’s degree in Computer Science and Mathematics from the Technical University of Munich. He is a recipient of the ETH Medal. His research aims to leverage foundation models for solving hard tasks through specialization and reinforcement learning. Furthermore, his work encompasses probabilistic inference, optimization, and online learning.

Event Info

Please help us plan ahead by registrating for the event via Meetup:
Event Registration .

What?	Test-Time Training Agents to Solve Challenging Problems
Who?	Jonas Huebotter, ETH Zurich
When?	November 5th 2025 @ 6 PM
Where?	Zoom
Registration	meetup event-site

---