The EPIA 2025 international conference
Keynote speakers
Paolo Torroni
Title of the talk: Argument Mining and Reasoning with Large Language Models
Abstract: The rapid evolution of Large Language Models has sparked discussions about their ability to reason and how they may affect human reasoning. In this talk, we look at LLM reasoning through the lens of argumentation. We will explore how argumentation theories can help investigate the limits of current LLMs and reason about their role in pubic debate.
Short bio: Paolo Torroni in an associate professor at the University of Bologna. He has collaborate in a large number of international projects, authoring over 180 articles in areas of artificial intelligence such as multi-agent systems, computational logics and natural language processing. His recent research focuses on NLP theory and applications, especially argument mining in legal texts and public debate. He is the former director of the Master’s Degree in Artificial Intelligence the head of the Language Technologies Lab of the University of Bologna.
Title of the talk: Model-based reinforcement learning and abstraction
Abstract: In reinforcement learning (RL), we develop techniques to learn to control complex systems, and over the last decade we have seen impressive successes ranging from beating grand masters in the game of Go, to real-world applications like chip design, power grid control, and drug design. However, nearly all applications of RL require access to an accurate and lightweight simulator from which huge numbers of trials can be sampled. In this talk, I will cover some settings where this is not the case, and where therefore we need to engage in some form of ‘model-based RL’ to learn an appropriate model.
Specifically, I will give an overview of a number of different problem settings (MDPs, POMDPs, and multiagent problems) and various corresponding approaches to learning and using models of the environment (ranging from deep-learning, to Bayesian inference, and from planning with MCTS variants to model-free RL), highlighting their strong points as well as limitations. Central to all these approaches is the notion of abstraction: how finely do we represent the world when learning and planning? And what impact might such abstractions actually have on theoretical guarantees of MBRL methods?
Short bio: Dr. Frans A. Oliehoek is Associate Professor at Delft University of Technology, where he is a leader of the sequential decision making group and director and co-founder of the ELLIS Unit Delft. He received his Ph.D. in Computer Science (2010) from the University of Amsterdam (UvA), and held positions at various universities including MIT, Maastricht University and the University of Liverpool. Frans’ research interests revolve around intelligent systems that learn about their environment via interaction, building on techniques from machine learning, AI and game theory. He has served as PC/SPC/AC at top-tier venues in AI and machine learning, and currently serves as associate editor for JAIR and AIJ. He is a Senior Member of AAAI, Fellow of ELLIS, and was awarded a number of personal research grants, including a prestigious ERC Starting Grant.
Frans A. Oliehoek
Mário Figueiredo
Title of the talk: The Why of AI: Causal Discovery from Observational Data
Abstract: In causal discovery, the aim is to uncover the causal mechanisms that drive the relationships between variables—a critical step beyond the correlational models prevalent in modern machine learning. This pursuit is foundational for the next generation of robust and explainable AI, with applications in most scientific fields. Although, in principle, identifying causal relationships requires interventions (experiments), it is often the case that this is impossible, impractical, or unethical. The challenge of learning cause-and-effect from purely observational data is therefore central to causal discovery. In this talk, after briefly surveying the field, I will discuss recent advances in causal discovery from data, namely the problem of distinguishing cause from effect on bivariate data.
Short bio: Mário Figueiredo received PhD and habilitation degrees in Electrical and Computer Engineering, both from Instituto Superior Técnico (IST), University of Lisbon, where he is now an IST Distinguished Professor and holder of the Feedzai Chair of Machine Learning. He is also a senior researcher and group leader at Instituto de Telecomunicações.
He received several honors, namely: Fellow of the Institute of Electrical and Electronics Engineers (IEEE), Fellow of the International Association for Pattern Recognition (IAPR), Fellow of the European Association for Signal Processing (EURASIP), Fellow of the European Laboratory for Learning and Intelligent Systems (ELLIS), member of the Portuguese Academy of Engineering and of the Lisbon Academy of Sciences.