Jonas Hübotter

I am a doctoral researcher in the Learning and Adaptive Systems Group at ETH Zurich working with Andreas Krause. Prior to this, I 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. As an intern at Citadel Securities, I have previously worked with Guillaume Basse and Sören Künzel on time-series prediction. I am a recipient of the ETH Medal.

My research aims to improve the performance of foundation models by utilizing tools from active learning for few-shot learning, active inference, and adaptive computation. Beyond this, I have broad interests including (approximate) probabilistic inference, optimization, and online learning.

Always feel free to reach out to me with things you find exciting.

Contacts: jhuebotter@ethz.ch Google Scholar GitHub Linkedin

Announcements

Sep, 2024	Our work Transductive Active Learning: Theory and Applications was accepted at NeurIPS 2024!
Jun, 2024	Our work on Transductive Active Learning with Application to Safe Bayesian Optimization was accepted as an oral presentation (top 5%) at the Workshop on Aligning RL Experimentalists and Theorists at ICML 2024.
Mar, 2024	Our work on Active Few-Shot Fine-Tuning was accepted at the Workshop on Bridging the Gap Between Practice and Theory in Deep Learning at ICLR 2024!

Selected Publications

Efficiently Learning at Test-Time: Active Fine-Tuning of LLMs

Jonas Hübotter, Sascha Bongni, Ido Hakimi , and 1 more author

arXiv preprint arXiv:2410.08020, 2024

Bib PDF

@article{hubotter2024efficiently,
  title = {Efficiently Learning at Test-Time: Active Fine-Tuning of LLMs},
  author = {H{\"u}botter, Jonas and Bongni, Sascha and Hakimi, Ido and Krause, Andreas},
  year = {2024},
  journal = {arXiv preprint arXiv:2410.08020},
}

Active Fine-Tuning of Generalist Policies

Marco Bagatella, Jonas Hübotter, Georg Martius , and 1 more author

arXiv preprint arXiv:2410.05026, 2024

Bib PDF Website

@article{bagatella2024active,
  title = {Active Fine-Tuning of Generalist Policies},
  author = {Bagatella, Marco and Hübotter, Jonas and Martius, Georg and Krause, Andreas},
  year = {2024},
  journal = {arXiv preprint arXiv:2410.05026},
}

NeurIPS Oral

Transductive Active Learning: Theory and Applications

Jonas Hübotter, Bhavya Sukhija, Lenart Treven , and 2 more authors

In NeurIPS , 2024

Oral Presentation at ICML Workshop on Aligning Reinforcement Learning Experimentalists and Theorists, 2024.

Bib PDF Code

@inproceedings{hubotter2024transductive,
  title = {Transductive Active Learning: Theory and Applications},
  author = {H{\"u}botter, Jonas and Sukhija, Bhavya and Treven, Lenart and As, Yarden and Krause, Andreas},
  year = {2024},
  booktitle = {NeurIPS},
}

Talks

Invited Talk
Machine Learning and Modelling Seminar, Charles University Prague, Nov 2024.
Transductive Active Learning for Fine-Tuning Large (Language) Models.
Contributed Talk
ICML Workshop on Aligning Reinforcement Learning Experimentalists and Theorists, Vienna, Jul 2024.
Transductive Active Learning with Application to Safe Bayesian Optimization, slides.

Supervision

I have had the privilege of advising several BSc and MSc students during their theses and semester projects.

Nicolas Menet: Efficiently Estimating Gaussian Probability of Maximality (with Parnian Kassraie)
Sascha Bongni: Active Fine-Tuning of Large Language Models
Pablo Lahmann: Safe Control as Inference (with Yarden As)
Anh Duc Nguyen: Safe Bayesian Optimization without Regret

You can find a list of potential projects here. If you are interested in working with me, please reach out.

Fun Projects

ActiveFT: A PyTorch Library for Active Fine-Tuning

Efficiently fine-tune large neural networks by intelligent active data selection.

SOCO: A Rust Library for Smoothed Online Convex Optimization

Algorithms for online convex optimization with an associated cost for movement in the decision space. Useful for resource allocation, contextual sequence prediction, portfolio management, and object tracking.

Solutions to Algorithms Lab 2021

Solutions to a wide variety of competitive programming-type questions.

Plaain

A serverless web app to organize and stream media from anywhere.

bifolia

Website of a landscape architecture firm.