Advanced Algorithms and Optimization Models Supported by Mathematical Theory (OptimaAI)

Basic Information

The OptimaAI project focuses on the development of innovative artificial intelligence algorithms and optimization models grounded in modern mathematical theories. The aim of the project is to build energy-efficient and scalable solutions for the analysis of complex data in bioinformatics, natural language processing, visual tracking, and industrial optimization. A key component of the project is the strong synergy between applied AI technologies (deep learning, NLP, edge AI) and fundamental mathematical disciplines (combinatorics and discrete mathematics, number theory, and group theory).

The project is structured around four work packages: WP1 – Development of advanced bioinformatics tools for single-cell transcriptome analysis; WP2 – Development of AI models for large-scale datasets and intelligent real-time camera signal processing; WP3 – Deep learning for combinatorial optimization and graph theory; and WP4 – Solving mathematical problems applicable to optimization through heuristics and metaheuristics using theoretical, combinatorial, and discrete approaches.

Expected results include open-source software, high-impact scientific publications, and the transfer of results into applications relevant to healthcare, industry, and the digital transition.

Full title	Advanced Algorithms and Optimization Models Supported by Mathematical Theory – OptimaAI
Project type	Institutional research project of the Josip Juraj Strossmayer University of Osijek
Funding source	National Recovery and Resilience Plan 2021–2026 (source 581 – Recovery and Resilience Facility), Programme Agreement between the Ministry of Science, Education and Youth and the University of Osijek
Contract number	581-UNIOS-54
Beneficiary	School of Applied Mathematics and Informatics, Josip Juraj Strossmayer University of Osijek
Principal investigator	Assoc. Prof. Domagoj Matijević, PhD
Project value	EUR 194,999.70 (100 % grant)
Duration	1 October 2025 – 30 September 2029 (48 months)

Publications

Journal papers

1. I. Matić, Zelevinsky segment and strongly positive representations, Canadian Mathematical Bulletin (2026), accepted. doi:10.4153/S0008439526102781
2. T. Prusina, J. Benić, D. Matijević, D. Ševerdija, Multi-person 2D human pose estimation: a benchmark for real-time applications, Engineering Review 46 (2026). Presented at MOTSP 2026. Code: HPEBench2D
3. J. Maltar, I. Marković, D. Matijević, I. Petrović, Deep recurrent visual place recognition based on softmax fine-tuned image representations, Engineering Review 46 (2026). Presented at MOTSP 2026.
4. M. Jukić Bokun, I. Soldo, Triangular D(−1)-tuples, Bull. Math. Soc. Sci. Math. Roumanie 69/2 (2026), 233–239.
5. D. Krupić, D. Matijević, N. Šuvak, J. Maltar, D. Ševerdija, Evaluating the agreement between human preferences, GPT-4V and Gemini Pro Vision assessments: Can AI recognize what people might like?, Computers in Human Behavior: Artificial Humans 6 (2025)

Conference proceedings papers

1. J. Sabljo, M. Đumić, M. Đurasević, Evolving Dispatching Rules for the Unrelated Parallel Machines Scheduling Problem with Precedence and Resource Constraints with Genetic Programming, GECCO ’26 – Genetic and Evolutionary Computation Conference, San Jose, Costa Rica, 13–17 July 2026, ACM. doi:10.1145/3795095.3805192
2. J. Sabljo, M. Đumić, M. Đurasević, Analysis of Training Models for Evolving Heuristics in the Unrelated Parallel Machines Scheduling Problem with Additional Resources, 2026 MIPRO 49th ICT and Electronics Convention, Opatija, Croatia, 2026. Code: UPMSP-PR
3. L. Borozan, B. Borozan, D. Ševerdija, Efficient Checkpointing via Object Serialization in C# Applications, 2026 MIPRO 49th ICT and Electronics Convention, Opatija, Croatia, 2026. Code: CheckpointingSystem

Papers under review

1. K. Sabo, R. Scitovski, D. Grahovac, Š. Ungar, Mahalanobis clustering for color image segmentation and application to skin lesion classification, Advances in Data Analysis and Classification (submitted)
2. K. K. Gottwald, T. Hofmann, S. Majstorović Ergotić, On the total inverse flow of a graph, Aequationes Mathematicae (submitted 2026)
3. S. Majstorović Ergotić, K. K. Gottwald, The total inverse flow of a graph: A Šoltés-type problem, Bulletin of the Malaysian Mathematical Sciences Society (submitted 2026)
4. L. Borozan, D. Matijević, Scenario-Driven Control of Real-Time LLM Agents, MIPRO Robotics 2026 – 5th Robotics Conference (RTA), Zadar, Croatia (accepted, proceedings forthcoming)

All papers acknowledge the OptimaAI project (581-UNIOS-54), funded by the National Recovery and Resilience Plan.

Conferences and workshops

Date	Event	Participation
25–28 Sep 2026	GCB 2026 – German Conference on Bioinformatics	B. Borozan, poster “Combinatorial CAR T-cell Circuit Design using Single- and Multi-Objective Local Search”
13–17 Jul 2026	GECCO 2026, San Jose, Costa Rica	J. Sabljo, paper presentation (Sabljo, Đumić, Đurasević); M. Đumić, attendance
10–12 Jul 2026	GECCO 2026 SIGEVO Summer School, San Jose	J. Sabljo
2026	MIPRO Robotics 2026 – 5th Robotics Conference (RTA), Zadar, Croatia	L. Borozan, paper presentation (Borozan, Matijević)
26–28 May 2026	MOTSP 2026 – Management of Technology Step to Sustainable Production, Zadar, Croatia	talks: J. Maltar (visual place recognition), J. Benić (2D pose estimation benchmark), T. Prusina (edge-based bearing condition monitoring); D. Ševerdija, attendance
25–29 May 2026	MIPRO 2026 – 49th ICT and Electronics Convention, Opatija, Croatia	talks: J. Sabljo (training models for evolving heuristics), L. Borozan (efficient checkpointing)
26 May 2026	Project workshop “Hyper-heuristic methods for electric vehicle routing problems”, FER, University of Zagreb	M. Đumić, R. Čorić, J. Sabljo

Open-source code repositories

Repository	Content	Work package
HPEBench2D	real-time multi-person 2D pose estimation benchmark	WP2
CheckpointingSystem	efficient checkpointing via object serialization in C# applications	WP3
UPMSP-PR	genetic programming for unrelated parallel machine scheduling	WP4

Collaborations and mobility

Institution	Collaboration	Period
University of Regensburg, Faculty of Informatics and Data Science (Germany)	research stay of Bartol Borozan; joint work on marker gene selection (WP1)	1 Oct 2025 – 30 Jun 2026
University of Regensburg, Faculty of Informatics and Data Science (Germany)	short research visit of Luka Borozan (Erasmus+)	24–28 Mar 2026
Chemnitz University of Technology, Faculty of Mathematics (Germany)	S. Majstorović Ergotić is scientific co-supervisor of PhD candidate Kurt Klement Gottwald; two joint papers (WP3)	since 2026
Faculty of Electrical Engineering and Computing, University of Zagreb	joint papers with M. Đurasević (WP4); project workshop, 26 May 2026	2026
GECCO 2026 SIGEVO Summer School, San Jose (Costa Rica)	training of Josipa Sabljo	10–12 Jul 2026

Research infrastructure

The project procured low-power edge AI devices (November 2025), which are used to validate compressed deep-learning models and to measure speed and resource consumption on constrained hardware (WP2, WP3).

In the second project year a next-generation GPU cluster is being procured; it will be integrated into the School’s HPC infrastructure and shared by all work packages (public procurement launched in October 2026). Equipment purchased with project funds is labelled with the EU emblem and funding statement.

News

• October 2026GPU cluster procurement launched; new team member

  Public procurement of a next-generation GPU cluster for the project has been launched; the cluster will be integrated into the School’s HPC infrastructure. From 1 November 2026, Ivo Šušac joins the team in WP4. The annual report for the first project year was submitted to the University.

• September 2026GCB 2026

  Bartol Borozan presented a poster on combinatorial CAR T-cell circuit design using local search (WP1) at GCB 2026 (German Conference on Bioinformatics).

• July 2026GECCO 2026, San Jose

  Josipa Sabljo presented a paper on evolving dispatching rules with genetic programming (WP4) at GECCO 2026 and attended the SIGEVO Summer School.

• May 2026MOTSP, MIPRO and a workshop at FER

  Project results were presented at MOTSP 2026 (Zadar; three WP2 talks) and MIPRO 2026 (Opatija; two talks from WP3 and WP4), and team members took part in a workshop on hyper-heuristic methods at FER Zagreb. Three public code repositories were released.

• March 2026Visit to the University of Regensburg; semi-annual report

  Luka Borozan made a short research visit (Erasmus+) to the Faculty of Informatics and Data Science, University of Regensburg, where Bartol Borozan is on a nine-month research stay. The semi-annual project report was submitted to the University.

• November 2025Edge AI devices procured

  Low-power edge AI devices were procured and put into operation; they are used to validate compressed deep-learning models and to measure real-time speed and resource consumption (WP2, WP3).

• October 2025Project start

  On 1 October 2025 the four-year OptimaAI project started, funded by the National Recovery and Resilience Plan (contract 581-UNIOS-54, EUR 194,999.70). The project brings together 15 researchers of the School in four work packages.

Funded by the European Union – NextGenerationEU. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Commission. Neither the European Union nor the European Commission can be held responsible for them.