Onat Özdemir

Informatics Forum

10 Crichton Street

Edinburgh EH8 9AB, UK

I am a PhD student at the School of Informatics, University of Edinburgh and currently a research intern at Five AI in Cambridge, working on vision–language–action models for autonomous driving.

My research interests lie in interpretability and representation analysis for multimodal models: what these models actually encode, how they arrive at a prediction, and when they should not be trusted. I have also worked on uncertainty and out-of-distribution detection in image classification.

Before Edinburgh I was an Applied Scientist at RadiusAI (2022–2025), where I built VLM-based visual search and multi-camera systems for retail environments. I completed my BSc and MSc in Computer Engineering at Middle East Technical University (METU), where my MSc thesis focused on interpretable zero-shot classification using concept bottleneck models.

I am always happy to talk about interpretability, multimodal representations, and embodied vision–language models.

news

Sep 28, 2026	Started as a Research Scientist Intern at Five AI, working on vision-language-action (VLA) models.
Aug 05, 2026	🎉 A Systematic Comparison of Training Objectives for Out-of-Distribution Detection in Image Classification accepted at UnCV @ ECCV 2026.
Feb 21, 2026	🎉 Explaining CLIP Zero-shot Predictions Through Concepts accepted to CVPR 2026! Project page: oonat.github.io/ezpc.
Sep 08, 2025	Started my PhD at the School of Informatics, University of Edinburgh.
Aug 05, 2025	Completed MSc in Computer Engineering at Middle East Technical University (METU), with a thesis on interpretable zero-shot classification with concept bottleneck models.

selected publications

CVPR

Explaining CLIP Zero-shot Predictions Through Concepts

Onat Ozdemir, Anders Christensen, Stephan Alaniz, and 2 more authors

In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Jun 2026

Bib PDF Website

@inproceedings{Ozdemir_2026_CVPR,
  author = {Ozdemir, Onat and Christensen, Anders and Alaniz, Stephan and Akata, Zeynep and Akbas, Emre},
  title = {Explaining CLIP Zero-shot Predictions Through Concepts},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  month = jun,
  year = {2026},
  pages = {31336-31345},
}

UnCV @ ECCV

A Systematic Comparison of Training Objectives for Out-of-Distribution Detection in Image Classification

Furkan Genc, Onat Ozdemir, and Emre Akbas

arXiv preprint arXiv:2603.07571, Mar 2026

Bib PDF

@article{gencc2026systematic,
  author = {Genc, Furkan and Ozdemir, Onat and Akbas, Emre},
  title = {A Systematic Comparison of Training Objectives for Out-of-Distribution Detection in Image Classification},
  journal = {arXiv preprint arXiv:2603.07571},
  month = mar,
  year = {2026},
}

MSc Thesis

Interpretable zero-shot classification with concept bottleneck models

Onat Ozdemir

Middle East Technical University, Aug 2025

Bib PDF

@mastersthesis{Ozdemir_2025_MSc,
  author = {Ozdemir, Onat},
  title = {Interpretable zero-shot classification with concept bottleneck models},
  school = {Middle East Technical University},
  month = aug,
  year = {2025},
}