Keynote speakers

A journey into pattern mining with FCA in practice

Mehdi Kaytoue

Mehdi Kaytoue is an Associate Professor of Computer Science at the Institut National des Sciences Appliquées de Lyon (INSA Lyon / LIRIS CNRS laboratory, France) since 2012. His academic journey brought him to Nancy – France (PhD at Université de Lorraine, 2011, entitled /Mining numerical data with formal concept analysis and pattern structures/), Belo Horizonte – Brazil (Universidade Federal de Minas Gerais, 2011), Lyon – France (post doc at INSA Lyon, 2012), to finally obtain his French Habilitation entitled /Contributions to Pattern Discovery and Formal Concept Analysis/in 2020.

His main research interests are artificial intelligence, data-mining, and especially pattern mining with formal concept analysis. Original methods and algorithms developed as part of his research have been applied to various domains with a particular attention to software engineering, digitalization, olfaction in neuroscience, geo-located social media analysis, electronic sports and video game analytics. This particular attention for applications lead him to conduct several projects with industries at different scales, such as the EU GRAISearch project as a Marie Curie fellow from 2014 to 2016 in Dublin – Ireland. He also took a sabbatical leave from 2018 to 2025 in the software engineering industry as a director of innovation. He co-advised six PhD thesis at the intersection of pattern mining, subgroup discovery and formal concept analysis. He was also a member of the scientific committee of the French National Research Council for Artificial Intelligence (ANR CE23 2023-2024). He (/et al./) is mostly known for his work on characterizing and mining numerical data with pattern structures (intervals, biclusters, implications, functional dependencies) and on video game and esports analytics.

Mehdi Kaytoue is involved in the FCA community for almost 20 years. He served as a program co-chair for ICFCA 2014 held in Cluj-Napoca, as an organization co-chair for CLA 2011 and acting every year as a reviewer for conferences and journals of the field (such as ICFCA, CLA) but also in data-mining and machine learning (Data mining and knowledge discovery journal, ECML/PKDD, ICDM, Machine Learning Journal, etc). He was also part of the ICFCA steering committee from 2014 to 2022.

One component of data science, Knowledge Discovery in Databases (KDD), deals in particular with the Data-Information-Knowledge pipeline with the aim of explaining models, relationships, or discovering hidden properties from data. Opposed to a purely statistical approach, a family of methods has met important success over the last thirty years: data mining and especially pattern mining.Their goal is to describe, summarize, and raise hypotheses from data.

In particular, pattern mining makes it possible to efficiently find regularities of various types (such as frequent patterns in a set of transactions, molecular sub-graphs characteristic of toxicity, locally co-expressed gene groups, etc.). In fact, where conventional approaches aim to validate or invalidate a hypothesis given a priori, the search for patterns is seen as an enumeration technique of all possible hypotheses (a set of exponential size with respect to the input data) verifying some given constraints or maximizing a certain interest for the expert. Once discovered, the best hypotheses can then be tested, validated or invalidated, and ultimately validated as a knowledge unit.

Formal Concept Analysis is a mathematical framework of excellence for KDD with pattern mining. This talk is not a tutorial on pattern mining with Formal Concept Analysis. Instead, it will discuss my experience over the last twenty years, focusing on making FCA practical through collaborations with other scientists, industrial partners, and during my recent time in the industry. We will explore three main axes: the formalism framing the data (e.g., binary, numerical, sequential, etc.) and pattern space (itemsets, data dependencies, biclusters, etc.), the methodological and algorithmic aspects, and finally, Knowledge Discovery ‘in practice’ through several concrete applications in fields such as microbiology, neuroscience, social networks, video game analytics, and software engineering.

Combining Concepts: Integrating Logical and Cognitive Theories of Concepts

Guendalina Righetti

Guendalina Righetti is a Postdoctoral Researcher at the University of Oslo, working in the ERC-funded project Construction in the Formal Sciences at the Department of Philosophy. Her research lies at the intersection of knowledge representation, logic, cognitive science, and applied ontology. She holds a PhD in Computer Science from the Free University of Bozen-Bolzano, where her research focused on modeling cognitive theories of concepts and conceptual combination within Description Logics.

Her work explores the integration of knowledge representation with insights from cognitive science, with the aim of advancing human-aligned and interpretable AI.

How do human minds represent and combine concepts? This question has been central in cognitive and experimental psychology, leading to the development of rich models of conceptual representation. Yet, these models often remain informal or underspecified, making them difficult to translate into precise computational frameworks. In contrast, formal approaches in Artificial Intelligence and Knowledge Representation typically prioritise efficiency and tractability, often at the expense of cognitive plausibility.

This talk presents an approach to modelling concepts and their combinations that aims to align better with findings from cognitive science. Drawing on psychological theories of concepts, we introduce Perceptron Logic — a formal framework that combines logical reasoning with techniques from statistical learning. This logic supports more interpretable and human-aligned representations, while remaining suitable for computational use. Based on this framework, we present an algorithm for concept combination grounded in the experimental research in cognitive psychology, and report on an empirical evaluation of its performance.

Interdisciplinary AI for Human Machine Interaction

Madalina Croitoru

Prof. Madalina Croitoru completed her PhD in Computer Science at the University of Aberdeen, Scotland, in 2007. In 2008, after two years of postdoctoral research at the University of Southampton, she joined the University of Montpellier, where she has been a full professor since 2019. Her broader research interests include human-AI interaction, with a specific focus on computational dialogues (e.g., deliberation, negotiation, and inquiry). Her research is carried out in the LIRMM laboratory of Montpellier in the IDH Robotics team, She is also currently pursuing a second PhD in cognitive sciences and psychology at the Paul Valery University in Montpellier investigating how elementary school children (8-12 yer olds) form attachment bonds with AI powered machines.

Ganesh Gowrishankar

Ganesh Gowrishankar is a Directeur de Recherche (Senior Researcher) with the Centre National de la Recherché Scientifique (CNRS), France and is currently located at the the Laboratoire d’Informatique, de Robotique et de Microélectronique de Montpellier (LIRMM) in Montpellier. He is a visiting researcher at the University of Tokyo, National Institute of Advanced Industrial Science and Technology (AIST) in Tsukuba, and University of Electro-communication in Tokyo. His research focuses on human-machine interactions by integrating core research in human sensori-motor control and cognitive neuroscience with robot control, learning and AI. Ganesh received his Bachelor of Engineering (first-class, Hons.) degree from the Delhi College of Engineering, India, in 2002 and his Master of Engineering from the National University of Singapore, in 2005, both in Mechanical Engineering. He received his Ph.D. in Bioengineering from Imperial College London, U.K., in 2010.