Research Publications and Software
Publication highlights compiled from ORCID, OpenAlex, and DBLP cross-checks.
- ORCID: 0000-0003-4292-989X
- OpenAlex: A5078324737
- DBLP: Marko Kokol
- Frontiers Loop: Profile
Slovenian Language Models: Cvetka and Gizela
My Cvetka and Gizela models hold first and second place on both the Slovene SuperGLUE and SloPragEval leaderboards on SloBench, as of October 27, 2026. Slovene SuperGLUE evaluates language understanding across multiple tasks; SloPragEval tests understanding of implied meaning in Slovenian.
| Benchmark | Model submission | Score | Rank |
|---|---|---|---|
| Slovene SuperGLUE | Cvetka-27B-tiv | 82.90% average | 1 |
| Slovene SuperGLUE | Gizela-31b-it-vit-think | 79.87% average | 2 |
| SloPragEval | Cvetka-27B | 88.75% average accuracy | 1 |
| SloPragEval | Gizela-31b-it-tav | 87.50% average accuracy | 2 |
Rankings refer to the published submissions on these two benchmarks at the date checked. Explore the current SloBench results and my models on Hugging Face.
Research Software by Marko Kokol
Tabnetics
Research software · Version 1.5.0 · Released August 3, 2026
HDLSS-focused tabular learning toolkit with distribution-aware preprocessing, portfolio feature selection, and game-theoretic method aggregation.
Designed for high-dimensional, low-sample-size tabular classification in bioinformatics and related structured-data research settings.
- Install:
pip install tabnetics - PyPI: https://pypi.org/project/tabnetics/
- Source: https://github.com/klokedm/tabnetics-public