diochnos/home

I am an Assistant Professor at the School of Computer Science at the University of Oklahoma. I am primarily interested in designing and analyzing machine learning algorithms with rigorous guarantees.

In particular, I have done work on the statistical and computational efficiency of supervised learning algorithms providing complexity or empirical bounds, or computational hardness results, within the context of adversarial supervised learning (different noise models, poisoning attacks, adversarial examples), randomized and local-search heuristic methods (evolvability), multiple-instance learning, and imbalanced data.

During the last couple of years my students and I are investigating semi-supervised learning, learning with streaming data, regularization methods, and related topics.

At the moment, the site is best viewed with Firefox / Iceweasel, Safari, Opera, and Chrome. You are using Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; ClaudeBot/1.0; +claudebot@anthropic.com).

Recent Work

My most recent paper is A Review of Pseudo-Labeling for Computer Vision.
This is joint work with my PhD student Jay Rothenberger and our collaborators from the University of Edinburgh Patrick Kage and Pavlos Andreadis.
The paper has been accepted for publication at the Journal of Artificial Intelligence Research (JAIR).
Please note that this is currently a pre-print and we intend to make some small changes so that we can align better with the final remarks of the reviewers as well as to include one repository that we identified after the submission to the journal.

Another recent paper is Dimensionally Reduced Open-World Clustering: DROWCULA.
This is joint work with a collaborating undergraduate student, Erencem Özbey.
Erencem was a visiting undergraduate student at OU during the fall of 2024 and worked in this topic.
The paper has been accepted and will appear at the Australasian Joint Conference on Artificial Intelligence (AJCAI), 2025.

A little bit further back (in July) this year, the following paper was accepted:
Meta Co-Training: Two Views are Better than One.
This is joint work with my PhD student Jay Rothenberger.
The paper has been accepted and will appear at the European Conference on Artificial Intelligence (ECAI), 2025.
🥇 Meta Co-Training is a semi-supervised learning method that achieves State-of-the-Art performance on ImageNet-10% and does very well on ImageNet-1% too.

Last year I had the following papers:

News