Development and validation of machine learning methods using prebuilt datasets

▌ Project lead

| Povilas Treigys
ResearchGate | VU MII profile

▌ Project overview

This project encompasses two primary components: infrastructure development and research activities. The project aims to establish an infrastructure that enables efficient and secure collection, storage, and sharing of large-scale datasets. To support these objectives, interdisciplinary research teams are systematically organised to foster collaboration across diverse domains. The research focuses on the development, adaptation, and validation of machine learning methods applied to heterogeneous datasets, aiming to achieve high accuracy, reliability, and applicability in real-world scenarios. Particular emphasis is placed on methodological universality and the capability to process both structured and unstructured data effectively, with applications spanning energy systems, biomedical analysis, and natural language processing.

▌ Project team

| Gražina Korvel ResearchGate | VU MII profile | Jolita Bernatavičienė ResearchGate | VU MII profile  | Gintautas Tamulevičius ResearchGate | VU MII profile

| Aurimas Rukšys LinkedIn | Eduardas Kutka LinkedIn | Rolandas Naujikas LinkedIn post

▌ Publications

  • Čiapas, Bernardas; Treigys, Povilas.
    Centre-loss—A preferred class verification approach over sample-to-sample in self-checkout products datasets. IET Computer Vision, 2024. DOI: 10.1049/cvi2.12302

  • Virbukaitė, S.; Bernatavičienė, J.; Imbrasienė, D.
    Glaucoma Identification Using Convolutional Neural Networks Ensemble for Optic Disc and Cup Segmentation. IEEE Access, 2024.

  • Vaitulevičius, A.; Bernatavičienė, J.; Markevičiūtė, J.; Naruševičiūtė, I.; Trakymas, M.; Treigys, P.
    Advancements in prostate zone segmentation: integrating attention mechanisms into the nnU-Net framework. Machine Learning: Science and Technology, 2024. DOI: 10.1088/2632-2153/ad7f24

  • Malūkaitė, J.; Treigys, P.; Bernatavičienė, J.
    Enhancing arrhythmia detection using an ensemble of transformer models for heartbeat classification. In: Machine Learning, Optimization, and Data Science, Springer, Cham, 2025, pp. 312–322. DOI: 10.1007/978-3-031-82484-5_23

  • Ramonaitė, J.; Gore, P.; Korvel, G.; Tamulevičius, G.
    Selection of preprocessing parameters for Wave-U-Net-based speech-denoising. In: Data Science in Applications: Towards AI-Driven Approaches, Springer, Cham, 2025, pp. 107–124. https://link.springer.com/chapter/10.1007/978-3-031-88486-3_5

  • Voveris, L.; Bernatavičienė, J.
    Saulės elektrinių matavimų klasifikavimas remiantis NLP ir mašininiu mokymusi. Vilnius University Open Series (Conference: “Lietuvos magistrantų informatikos ir IT tyrimai”, 2025). https://www.journals.vu.lt/open-series/en/article/view/41956/39348

  • Valatka, M.
    Neuroniniais tinklais grįstų triukšmo šalinimo EKG signale metodų eksperimentinis tyrimas. Vilnius University Open Series (Conference: “Lietuvos magistrantų informatikos ir IT tyrimai”, 2025). https://www.journals.vu.lt/open-series/en/article/view/41953/39345

  • Žygas, A.; Korvel, G.
    Muzikos garso šaltinių atskyrimo giliojo mokymosi modelio SCNet apmokymas skirtingais duomenų rinkiniais. Vilnius University Open Series (Conference: “Lietuvos magistrantų informatikos ir IT tyrimai”, 2025). https://www.journals.vu.lt/open-series/en/article/view/41962/39353

▌ Scientific dissemination & public outreach

  • Dastgeer, Sobia; Treigys, Povilas.
    Transforming black-box models into explainable AI for breast cancer recognition.
    DAMSS: 15th conference on data analysis methods for software systems, Druskininkai, Lithuania, November 28–30, 2024. DOI: 10.15388/DAMSS.15.2024

  • Karina Čiurlienė.
    Analysis of Event and Human Factor-Based Decision-Making in Cybersecurity Exercises Using MCDM.
    DAMSS 2024. DOI: 10.15388/DAMSS.15.2024

  • Urtė Deinoravičiūtė; Jurgita Markevičiūtė; Povilas Treigys.
    Cointegration of Functional Time Series.
    DAMSS 2024. DOI: 10.15388/DAMSS.15.2024

  • Article: VU Mathematics and Informatics Faculty Data Center: From Big Data to a Breakthrough in Quantum Computing.
    IQ magazine, December 2025 (No. 12 (189)); online: IQ article link

  • Preparation of the official Lithuanian Quantum Technologies Roadmap (Quantum Agenda) on 28 February 2025:
    English version | Lithuanian version

▌ Contribution to programme objectives & external funding

The project contributes to programme objectives by strengthening R&D quality through collaborations, infrastructure, and funding leverage.

  • A collaboration agreement between MIF and the Poznań Supercomputing and Networking Center (PCSS) in quantum technologies enables best‑practice exchange and access to quantum computing resources.

  • Based on the deployed infrastructure, an automated platform was designed and implemented for the long‑term storage and efficient management of scientific datasets. The system ensures data availability, security, and integrity, provides a browser-based user interface for exploring datasets and monitoring their status, and exposes an API for integration with other information systems and for automated operations. To guarantee data consistency and interoperability, a standardised structure was defined with mandatory directories: data/ for datasets and meta/ for metadata (including a metadata.xml description). Source code (public repository):

  • Project results contributed to the successful funding of the Artificial Intelligence Factory application. Vilnius University is the coordinating institution. Project value — €130 million EUR.