Peter Sinčák
Professor of Artificial Intelligence
Institute of Artificial Intelligence
Technical University of Košice (TUKE)
Slovakia, European Union
Biography
Peter Sinčák is a Professor of Artificial Intelligence at the Technical University of Košice (TUKE), Slovakia, European Union. He received his MSc from TUKE in 1984 and his PhD from the Czech Academy of Sciences in Prague in 1992, and became a full professor in 2001.
In 1993, he pioneered neural network education in Slovakia by introducing the country's first university course dedicated to neural networks. He is a former Head of the Department of Cybernetics and Artificial Intelligence and has extensive experience in developing and leading AI education at bachelor's, master's and doctoral levels.
He has authored or co-authored more than 100 scientific publications and two monographs, and has edited books for international academic publishers. His research encompasses neural networks, computational intelligence, cloud-based AI and human-robot interaction. In recent years, his work has focused on diverse neural network architectures, including large-scale models for multimodal data processing, with particular attention to their adaptation and experimental evaluation.
Prof. Sinčák has contributed to national and international research projects, including the European initiatives AI4EU, LIFEBOTS and ROBOCOM++. His international experience includes a Marie Curie fellowship at Siemens in Vienna and numerous invited talks worldwide, including in China, Japan, South Korea, the United States, Italy and Hungary. He maintains longstanding research collaborations in Asia and actively promotes university-industry cooperation.
As an emeritus professor, he is committed to fostering mutual understanding, international scientific collaboration and cultural exchange, guided by the conviction that responsible AI should strengthen human capabilities and serve humanity.
Quo Vadis, Open Multimodal Neural Networks? Architectures, Adaptation and Experimental Perspectives
Multimodal models are an important component of contemporary artificial intelligence and are likely to play an increasingly significant role in its future development. This talk examines their architectural principles and practical adaptation through selected student research projects at the Technical University of Košice.
The presentation explores four vision-language models — CLIP, BLIP-2, BLIP-3 and OpenFlamingo — alongside two multimodal diffusion models, LTX-2 and JavisDiT++. Their architectures illustrate contrasting approaches to integrating modalities for understanding and generation.
Experiments cover image-text retrieval, visual question answering, multilingual captioning and domain-specific video generation. The studies investigate few-shot prompting, modular processing pipelines and several forms of model fine-tuning. Computationally demanding experiments use TUKE's PERUN HPC infrastructure, equipped with NVIDIA H200 GPUs.
The discussion considers model selection, reproducible experimentation, university research, cloud-based and local deployment, multimodal human-computer interaction, and the responsibility of researchers, developers and users to ensure that AI supports human thinking, critical judgement and intellectual autonomy.
Research Interests
- Neural Networks
- Computational Intelligence
- Multimodal Artificial Intelligence
- Large-scale Neural Network Architectures
- Cloud-based AI
- Human-Robot Interaction
- Model Adaptation and Experimental Evaluation
Selected Highlights
- Introduced Slovakia's first university course dedicated to neural networks in 1993
- More than 100 scientific publications
- Author or co-author of two monographs
- Marie Curie Fellowship at Siemens in Vienna
- Contributor to AI4EU, LIFEBOTS and ROBOCOM++
- Invited talks across Asia, Europe and the United States
- Longstanding international research collaborations in Asia