 |
Prof. Konstantinos Diamantaras
Dr. Konstantinos Diamantaras is Vice Rector of the International Hellenic University and Professor at the Department of Information and Electronics Engineering. He brings over thirty-five years of experience in Machine Learning and Artificial Intelligence. He holds a Diploma from the National Technical University of Athens and a Ph.D. in Electrical Engineering from Princeton University. His research, which includes over 180 publications, focuses on machine learning and its applications in many fields, including natural language processing, medical informatics, wireless communications, recommendation systems, economics, educational data mining, ea. Dr. Diamantaras has held leadership roles such as Chairman of the IEEE Machine Learning for Signal Processing Technical Committee and served as an Associate Editor for various IEEE journals. He is listed in the Stanford University/Elsevier Top 2% Scientists. His long-standing commitment to these fields has significantly contributed to both academic and applied research in AI and ML. |
Talk title: Mitigating Hallucinations in Large Language Models
Abstract: Large Language Models (LLMs) have transformed natural language generation, yet they continue to suffer from hallucinations—outputs that are linguistically fluent but factually incorrect, ungrounded, or logically inconsistent. This talk offers a comprehensive synthesis of state-of-the-art hallucination mitigation techniques, organized into six methodological categories: Training and Learning Approaches, Architectural Modifications, Input and Prompt Optimization, Post-Generation Quality Control, Interpretability and Diagnostic Methods, and Agent-Based Orchestration. The presentation will examine the underlying causes of hallucinations—from noisy or outdated training data to probabilistic decoding biases—and review cross-stage mitigation strategies spanning pre-generation, generation, and post-generation phases. Emphasis will be placed on hybrid paradigms such as retrieval-augmented generation, contrastive and reinforcement learning from human or AI feedback, and emerging self-reflective multi-agent systems. The talk will address persistent challenges—such as scalability, latency–accuracy trade-offs, and benchmark inconsistencies—while highlighting underexplored research directions including knowledge-grounded fine-tuning, factual–creative balance, and context-aware verification pipelines. Ultimately, it will focus on designing more reliable, transparent, and context-sensitive LLMs for high-stakes domains such as healthcare, law, and defense.
 |
Asoc. Prof. Jakub Kudela
Jakub Kůdela is an Associate Professor at the Institute of Automation and Computer Science, Brno University of Technology (Brno, Czech Republic), where he has worked since 2018, progressing from Lecturer/Assistant to Assistant Professor (2020–2024) and, since 2024, Associate Professor following his habilitation in Computer Science and Informatics at the Faculty of Information Technology. He holds an Ing. degree in Mathematical Engineering from Brno University of Technology and a Ph.D. in Applied Mathematics (2019). His research focuses on optimization, mathematical programming, and surrogate-model- and heuristic-based methods for engineering applications. He has authored 64 peer-reviewed papers (38 indexed in Web of Science), including a 2022 article in Nature Machine Intelligence, and holds an h-index of 11 (WoS) / 14 (Google Scholar). He is an active reviewer for more than 90 papers in journals such as IEEE Transactions on Evolutionary Computation and Expert Systems with Applications, a Programme Committee member of GECCO and the IEEE Conference on Artificial Intelligence, and currently leads a Czech Science Foundation project on benchmarking global optimization methods (2024–2026). He supervises four Ph.D. students and teaches courses in informatics and optimization at Brno University of Technology. |
Talk title: A Tour of the Separate Islands of Model-based Optimization
Abstract: Expensive black-box optimization has produced several mature research traditions, including DIRECT-type partitioning methods, deterministic response-surface optimization, model-based derivative-free optimization, Bayesian optimization, and surrogate-assisted evolutionary algorithms. Although these approaches address closely related problems, they have developed largely within separate communities, with distinct terminology, software ecosystems, benchmark suites, and experimental conventions. This talk reviews their common principles, methodological differences, and current research directions, then examines the surprisingly limited evidence from direct cross-family comparisons.
 |
Prof. Ruslan Mitkov
Ruslan Mitkov is Professor at Lancaster University (UK) and Distinguished Professor at the University of Alicante (Spain). He previously led the Research Group in Computational Linguistics at the University of Wolverhampton, where he also served as Director of the Research Institute of Information and Language Processing. He has worked in Natural Language Processing, Computational Linguistics, Machine Translation and Translation Technology since the early 1980s, with seminal contributions to anaphora resolution, automatic test generation and translation memory systems. His recent research focuses on Deep Learning, Large Language Models and AI applications in NLP and Translation Technology. He is author of more than 350 publications, including 20 books. He wrote Anaphora Resolution (Longman) and edited The Oxford Handbook of Computational Linguistics (Oxford University Press). He is Executive Editor of the journal Natural Language Processing (Cambridge University Press) and Editor-in-Chief of the Natural Language Processing book series at John Benjamins. Prof Mitkov has been keynote or invited speaker at more than 250 international conferences and has chaired over 80 conferences on NLP, Translation Technology and Applied Linguistics. He designed and directs the first Erasmus Mundus Master's programme in Technology for Translation and Interpreting. He is Vice President of AsLing, the International Association for promoting Language Technology, and has been awarded the title of Doctor Honoris Causa on three occasions. |
Talk title: Natural Language Processing in the Era of Artificial Intelligence: The Wind of Change is Blowing Stronger
Abstract: Natural Language Processing (NLP) is currently undergoing profound and unprecedented transformation. While it has long been acknowledged that NLP technologies are inherently imperfect and far from achieving complete accuracy, recent advances have significantly reshaped the field of language and translation technologies. First Deep Learning methods, and now Large Language Models (LLMs), have revolutionised the field and captured widespread attention across both academia and industry. This entertaining and accessible talk will shed light on the future of NLP in the era of Artificial Intelligence (AI). The talk will begin by outlining the history of NLP and Machine Translation, reviewing the advances driven by Deep Learning and LLMs. The speaker will then offer a critical evaluation of the use of LLMs in NLP and Machine Translation, focusing on his recent comparative studies involving LLMs, traditional Deep Learning architectures, and rule-based methodologies across selected NLP tasks and applications. Finally, the presentation will turn to the future of NLP. Generative AI is certainly continuing to improve but are there any attendant dangers? While the speaker makes no claim to prophetic insight, he will draw on his extensive experience and use linguistic intelligence as a benchmark to offer predictions on how artificial intelligence may evolve relative to human intelligence. The talk will also consider how language professions themselves are likely to evolve in the years ahead.