Jannik Kossen

Jannik Kossen

πŸ‘¨β€πŸ’» PhD Student in Machine Learning

πŸ₯£ OATML Oxford

πŸ‚ OxCSML

Hi! πŸ‘‹

Thanks for stopping by!

I am a PhD student at the University of Oxford supervised by Yarin Gal in OATML and Tom Rainforth in OxCSML. I am interested in uncertainties in large language models and computer vision, data-efficient machine learning, contrastive learning of vision-language models, (vaguely) Bayesian deep learning, active data selection, and deep non-parametric models. I received an MSc in Physics from Heidelberg University and have spent time studying in Bremen, Darmstadt, Padova, and at University College London.

I was a Student Researcher at Google Research, working on large scale contrastive learning, and a Research Scientist Intern at DeepMind, exploring active feature acquisition for temporal multimodal data.

I am interested in the societal and ethical implications of AI: I have co-authored a book explaining machine learning to a broad audience, discussed the ethics of AI at the Berlin-Brandenburg Academy of Sciences, and gathered real-world field experience at Bosch.

πŸ“° News

All News»

01/24 – Our work on how in-context learning in LLMs learns label relationships has been accepted to ICLR 2024.

09/23 – Work from my internship at Google on contrastive learning with pre-trained models has been accepted to NeurIPS 2023.

08/23 – New preprint on how in-context learning in large language models learns label relationships.

07/23 – Work from my internship at Google on contrastive learning with pre-trained models has been accepted at ES-FoMo Workshop at ICML.

07/23 – Work from my internship at DeepMind on Active Acquisition for Multimodal Temporal Data: A Challenging Decision-Making Task has been accepted at Transactions on Machine Learning Research.

05/23 – New Preprint vom my time as student rearcher at Google: Three Towers: Flexible Contrastive Learning with Pretrained Image Models.

11/22 – Our work on Active Surrogate Estimators: An Active Learning Approach to Label-Efficient Model Evaluation has been accepted as an oral to NeurIPS 2022.

11/22 – Work from my internship at DeepMind on Active Acquisition for Multimodal Temporal Data: A Challenging Decision-Making Task has been accepted at the Foundation Models for Decision Making NeurIPS 2022 Workshop.

πŸ“„ Selected Publications

Three Towers: Flexible Contrastive Learning with Pretrained Image Models.

Modelling Videos of Physically Interacting Objects.

πŸ“« Contact