TY - JOUR AU - Morvay, Balázs Tibor AU - Torma, Szabolcs AU - Pitrik, József AU - Szegletes, Luca TI - Enhancing multi-paradigm EEG signal classification in cross-subject settings using optimal transport JF - BIOMEDICAL SIGNAL PROCESSING AND CONTROL J2 - BIOMED SIGNAL PROCES VL - 113 PY - 2026 IS - Part B PG - 13 SN - 1746-8094 DO - 10.1016/j.bspc.2025.108892 UR - https://m2.mtmt.hu/api/publication/36388678 ID - 36388678 N1 - Funding Agency and Grant Number: European Union [RRF-2.3.1-21-2022-00004]; Doctoral Excellence Fellowship Programme - National Research Development and Innovation Fund of the Ministry of Culture and Innovation and the Budapest University of Technology and Economics; Frontline" Research Excellence Programme of the Hungarian National Research, Development and Innovation Office-NKFIH [KKP133827]; Momentum Program of the Hungarian Academy of Sciences [LP2021-15/2021] Funding text: Supported by the European Union project RRF-2.3.1-21-2022-00004 within the framework of the Artificial Intelligence National Laboratory. The project supported by the Doctoral Excellence Fellowship Programme (DCEP) is funded by the National Research Development and Innovation Fund of the Ministry of Culture and Innovation and the Budapest University of Technology and Economics. Pitrik was supported by the "Frontline" Research Excellence Programme of the Hungarian National Research, Development and Innovation Office-NKFIH (grant no. KKP133827) and by the Momentum Program of the Hungarian Academy of Sciences (grant no. LP2021-15/2021) ; The authors would like to thank Balint Beres for his valuable help during the preparation of this manuscript. Part number: B AB - Electroencephalography (EEG) signals have great variability between individuals, which makes the application of machine learning-based algorithms challenging in a cross-subject manner. In this work, a new approach is presented for EEG cross-subject classification, which combines feature extraction, classifiers and Wasserstein Barycenter Transport (WBT), a multi-source domain adaptation algorithm based on optimal transport, to reduce the distribution shift between subjects. The adaptation process is fully unsupervised, and corresponds to the offline scenario when a brain–computer interface is tuned on a new subject. The method is evaluated on three datasets, in four classification tasks, using four popular classification methods in a cross-subject setting. It achieves state-of-the-art 79.59% balanced accuracy in classifying visually evoked potential signals on the VEPESS dataset, surpassing the previous best by more than 10%. It also shows competitive results in three motor imagery classification tasks, reaching 53.97% average accuracy on the BCIC4D2a dataset, surpassing most optimal transport-based methods. The proposed method enhances classical machine learning models by introducing a novel, computationally efficient, and interpretable algorithm. This integration provides a straightforward path for clinical practitioners and researchers to augment their existing models without resorting to more complex and less transparent deep learning frameworks. LA - English DB - MTMT ER - TY - CHAP AU - Morvay, Balázs Tibor AU - Torma, Szabolcs AU - Béres, Bálint AU - Pitrik, József AU - Szegletes, Luca ED - Vajk, István ED - Dunaev, Dmitriy TI - Cross-subject EEG signal classification using optimal transport T2 - Proceedings of the Automation and Applied Computer Science Workshop 2025 PB - Budapest University of Technology and Economics, Department of Automation and Applied Informatics CY - Budapest SN - 9789634219989 PY - 2025 SP - 141 EP - 150 PG - 10 UR - https://m2.mtmt.hu/api/publication/36334173 ID - 36334173 LA - English DB - MTMT ER - TY - CHAP AU - Morvay, Balázs Tibor AU - Torma, Szabolcs AU - Béres, Bálint AU - Szegletes, Luca AU - Pitrik, József TI - An optimal transport-based domain adaptation method for enhancing cross-subject EEG emotion recognition T2 - 16th IEEE International Conference on Cognitive Infocommunications CogInfoCom 2025 PB - IEEE CY - Piscataway (NJ) SN - 9798350356939 T3 - Proceedings - International Conference on Cognitive Infocommunications, CogInfoCom, ISSN 2380-7350 PY - 2025 SP - 93 EP - 98 PG - 6 DO - 10.1109/CogInfoCom66819.2025.11200973 UR - https://m2.mtmt.hu/api/publication/36391585 ID - 36391585 N1 - Balazs Tibor Morvay, Szabolcs Torma, Balint Beres equally contributed Supported by the European Union project RRF-2.3.1-21-202200004 within the framework of the Artificial Intelligence National Laboratory. The project supported by the Doctoral Excellence Fellowship Programme (DCEP) is funded by the National Research Development and Innovation Fund of the Ministry of Culture and Innovation and the Budapest University of Technology and Economics. Pitrik was supported by the “Frontline” Research Excellence Programme of the Hungarian National Research, Development and Innovation Office - NKFIH (grant no. KKP133827) and by the Momentum Program of the Hungarian Academy of Sciences (grant no. LP2021-15/2021); AB - Electroencephalography has become a very popular, non-invasive method for brain signal measurement. However, it suffers from large inter-subject variabilities, which makes it challenging to perform analysis with good generalization. This work examines emotion recognition from EEG signals in a cross-subject setting. We identify this task as a domain adaptation problem and describe a method that combines feature engineering, the Wasserstein Barycenter Transport algorithm, and a support vector classifier to perform classification in this setting. This novel approach is fully unsupervised, based on optimal transportation, and achieves state-of-the-art results on the SEED-IV benchmark dataset under a leave-one-subject-out protocol. LA - English DB - MTMT ER - TY - CHAP AU - Morvay, Balázs Tibor AU - Szegletes, Luca ED - Dunaev, Dmitriy ED - Vajk, István TI - Improving latent fMRI features classification using transformations T2 - Proceedings of the Automation and Applied Computer Science Workshop 2024 : AACS'24 PB - Budapesti Műszaki Egyetem (BME) CY - Budapest SN - 9789634219606 PY - 2024 SP - 213 EP - 220 PG - 8 UR - https://m2.mtmt.hu/api/publication/35160336 ID - 35160336 LA - English DB - MTMT ER - TY - JOUR AU - Gulyás, Gábor AU - Kiss, Attila Csaba AU - Morvay, Balázs Tibor AU - Béres, Bálint AU - Várfalvi, Marianna AU - Rákóczi, Ildikó AU - Kovács, Edit AU - Papp, Csaba Sándor AU - Zsuga, Judit TI - Gerincferdülés szűrése mesterséges intelligenciával JF - EGÉSZSÉGÜGYI INNOVÁCIÓS SZEMLE J2 - EÜ INNOV SZLE VL - 3 PY - 2024 IS - 1 SP - 40 EP - 55 PG - 16 SN - 2939-6026 DO - 10.56626/r7hd9295 UR - https://m2.mtmt.hu/api/publication/35573931 ID - 35573931 AB - Early detection of scoliosis and other locomotor disorders is critical for effective treatment and improving the quality of life of those affected. The technology enables school nurses to make a more accurate diagnosis and make timely recommendations for the necessary interventions. In this study, we present the role of nurses in school health screening and the Tartás Pajtás tool, which was created to support school nurse screening. With the involvement of several nurses, we tested this device in a real environment and analyzed its possibilities and limitations. The results presented in the study were realized in the framework of the international project IML4E. LA - Hungarian DB - MTMT ER - TY - CHAP AU - Morvay, Balázs Tibor AU - Szegletes, Luca ED - Vajk, István ED - Dunaev, Dmitriy TI - Keypoint-keeping Super-Resolution T2 - Proceedings of the Automation and Applied Computer Science Workshop 2023 (AACS'23) PB - Budapesti Műszaki Egyetem, Automatizálási és Alkalmazott Informatikai Tanszék CY - Budapest SN - 9789634219262 PY - 2023 SP - 186 EP - 192 PG - 7 UR - https://m2.mtmt.hu/api/publication/34143812 ID - 34143812 LA - English DB - MTMT ER - TY - CHAP AU - Morvay, Balázs Tibor ED - Molnár, Dániel ED - Molnár, Dóra TI - Performance and Portability Analysis of Deep Learning Models on iOS Operating System T2 - XXVI. Tavaszi Szél Konferencia 2023 PB - Doktoranduszok Országos Szövetsége (DOSZ) CY - Budapest SN - 9786156457370 PY - 2023 SP - 407 EP - 412 PG - 6 UR - https://m2.mtmt.hu/api/publication/34531266 ID - 34531266 LA - English DB - MTMT ER - TY - CHAP AU - Morvay, Balázs Tibor AU - Béres, Bálint AU - Torma, Szabolcs AU - Szegletes, Luca TI - Diffusion probabilistic model based face anonymization in embedded environments T2 - 2023 14th IEEE International Conference on Cognitive Infocommunications (CogInfoCom) PB - IEEE Hungary Section CY - Piscataway (NJ) SN - 9798350325652 PY - 2023 SP - 135 EP - 140 PG - 6 DO - 10.1109/CogInfoCom59411.2023.10397492 UR - https://m2.mtmt.hu/api/publication/34183765 ID - 34183765 N1 - Conference code: 196901 Export Date: 4 March 2024 Funding text 1: The work presented in this paper has been carried out in the frame of project no. 2019-2.1.1-EUREKA-2020-00016, which has been implemented with the support provided by the National Research, Development, and Innovation Fund of Hungary, financed under the 2019-2.1.1-EUREKA funding scheme. LA - English DB - MTMT ER - TY - CHAP AU - Trautsch, László Kálmán AU - Morvay, Balázs Tibor AU - Csikós, Marcell ED - Molnár, Dániel ED - Molnár, Dóra TI - IDENTIFYING DRUG INTERACTIONS WITH A VARIATIONAL GRAPH AUTO-ENCODER USING MISSING DATA IMPUTATION T2 - XXVI. Tavaszi Szél Konferencia 2023 PB - Doktoranduszok Országos Szövetsége (DOSZ) CY - Budapest SN - 9786156457370 PY - 2023 SP - 374 EP - 379 PG - 6 UR - https://m2.mtmt.hu/api/publication/34457490 ID - 34457490 LA - English DB - MTMT ER - TY - CHAP AU - Béres, Bálint AU - Morvay, Balázs Tibor AU - Szegletes, Luca ED - Dunaev, Dmitriy ED - Vajk, István TI - Face Anonymization for Single Person Pose Estimation: A Comparative Study of the Effect of Face Anonymization T2 - Proceedings of the Automation and Applied Computer Science Workshop 2022 (AACS'22) PB - Budapesti Műszaki Egyetem, Automatizálási és Alkalmazott Informatikai Tanszék CY - Budapest SN - 9789634218753 PY - 2022 SP - 252 EP - 258 PG - 7 UR - https://m2.mtmt.hu/api/publication/34143868 ID - 34143868 AB - With the rising importance of data protection and individual privacy, face anonymization has drawn increased attention from the research community in recent years. Face anonymization is a computer vision technique where facial information is removed from digital images and videos. In this paper, we explore the effectiveness of different face anonymization methods for a specific use case: pose estimation. Pose estimation is widely used in augmented reality, health care and robotics. In human pose estimation a person’s movements are tracked by finding the location of selected keypoints. First, we anonymize the input image (remove facial information), then we apply a pose estimation algorithm. LA - English DB - MTMT ER -