Jorik Jooken (KU Leuven, Belgium): How can computers help in obtaining new results in graph theory?
This talk consists of two parts. In the first part, I will give an introduction to the domain of computer-assisted graph theory, which is concerned with developing algorithms in order to help researchers gain insights into various graph theoretical questions. In particular, I will discuss a broad range of popular techniques from this domain and briefly talk about their applications. In the second part, I will zoom in on an algorithm for exhaustively generating graphs and show how executing this algorithm leads to new insights in the domain of graph colouring.
This talk is based on two papers, available at https://arxiv.org/pdf/2508.20825 and https://arxiv.org/pdf/2404.11704.
Jorik Jooken is an FWO Postdoctoral Fellow at KU Leuven, working in the research group of Jan Goedgebeur. He earned his Ph.D. from KU Leuven in 2023, where he has been a postdoctoral researcher since his graduation. As of August 2026, he will join Leiden University in the Netherlands as an Assistant Professor.
Apart from his activities at KU Leuven, he has also spent several months in various universities abroad such as Comenius University (Slovakia), Nankai University (China), Western Sydney University (Australia) and Durham University (United Kingdom).
He is an expert in the field of discrete algorithms for combinatorial problems, with a particular emphasis on computer-assisted graph theory. He collaborates frequently with other mathematicians and computer scientists on various topics such as combinatorial optimization, extremal graph theory, graph coloring, long cycles in graphs, and enumeration of graphs. The topics of some of his latest publications include vertex-critical (P5,W4)-free graphs, (k,g)-graphs without (g+1)-cycles, girth and connectivity of cubic graphs with a unique longest cycle and a survey on computer-assisted graph theory.
He has been the advisor to three doctoral students and to numerous bachelor’s and master’s students.
Jernej Vičič (University of Primorska, Slovenia): AI-Generated Text Detection for Under-Resourced Languages
The rapid development of large language models (LLMs) has created a growing need for reliable methods to distinguish AI-generated text from human-written text. Existing detection approaches include statistical methods, stylometric analysis, perplexity-based techniques, transformer-based classifiers, and large neural models trained specifically for AI-text detection. Although many of these methods achieve high accuracy on benchmark datasets, their performance often decreases when applied to new domains, new language models, paraphrased texts, or modified content. Another important limitation is language bias. Most existing detectors are developed and evaluated mainly for English, while their performance is often much lower for other languages, especially for morphologically rich and low-resource languages. This talk investigates a simple question: how do existing detectors deal with under represented (and in the era of LLMs all non-English languages belong to this group) and morphologically rich languages.
The talk also proposes a simple, but not computationally cheap, and interpretable detection approach based on Benford's law and transformer embeddings. We hypothesize that AI-generated texts follow Benford's law more closely than human-written texts when numerical patterns are extracted from contextual embedding representations. We also hypothesize that the investigated method transfers well across languages and domains. To test this hypothesis, token-level embeddings are obtained from multilingual and language-specific transformer models and transformed into leading-digit distributions. The similarity between the observed distributions and Benford's law is measured using Kullback–Leibler divergence, chi-square statistics, mean squared error, and the coefficient of determination.
The proposed framework is evaluated on multilingual datasets, purposefully tailored for the ITAT conference, containing English, Slovene, Czech, and Slovak texts. The experiments examine whether the Benford-law hypothesis holds across languages, whether the approach can transfer between languages, how different embedding models affect performance. Both multilingual and language-specific transformer models are evaluated. A simple TF-IDF and Multinomial Naive Bayes classifier is used as a reference baseline along with the comparisons reported by the state of the art research.
The goal of this study is to determine whether Benford-law-based features capture general characteristics of AI-generated text that are less dependent on language than existing approaches. By focusing on simple statistical properties of embedding representations, the proposed method offers a more language-independent alternative for multilingual AI-text detection, particularly for languages that are underrepresented in current research.
Jernej Vičič is a Full Professor of Computer Science at the University of Primorska, where he is affiliated with the Faculty of Mathematics, Natural Sciences and Information Technologies (FAMNIT) and the Andrej Marušič Institute (IAM). He earned his B.Sc., M.Sc., and Ph.D. in Computer Science from the University of Ljubljana, He has also serves as Head of the Centre for Application Development at UP IAM and Head of the Distributed Ledger Technologies and Language Technologies Lab at UP FAMNIT.
Professor Vičič combines deep expertise in artificial intelligence and language technologies with extensive experience in distributed computing and large-scale data systems. His work bridges fundamental research and practical innovation, contributing to advances in machine translation, intelligent systems, decentralized infrastructures, and emerging AI technologies.
His recent research includes innovative work on occupancy estimation from indoor air-quality data, advanced retrieval-augmented generation methods using hypothetical prompt embeddings, large-scale detection of wash trading in NFT markets, distributed frameworks for large language model inference, open-source AI ecosystems, privacy-preserving wireless sensor networks, anonymous routing protocols, and decentralized multiplayer architectures. These contributions demonstrate a consistent focus on scalable intelligent systems, privacy-aware computing, and the practical deployment of advanced AI technologies.
Professor Vičič has supervised three completed doctoral students and more than sixty master's and bachelor's students.
