Leiden ↔ Chile International Research Seminar on Computer Science and AI

My research sits at the intersection of automated machine learning (AutoML) and open science, with a focus on making AI systems more powerful, transparent, and trustworthy. I work on AutoML methods that take over labor-intensive tasks in the machine learning pipeline (such as hyperparameter optimisation and neural architecture search), freeing researchers and practitioners to focus on higher-level concerns like model fairness, interpretability, and reliability.

A core belief driving my work is that automation should empower humans in the machine learning loop, not replace their judgment. By reducing technical friction, AutoML lowers barriers to entry and enables more thoughtful, human-centred AI development.

I am a co-founder and maintainer of OpenML (openml.org), an open platform for sharing machine learning experiments, datasets, and workflows. OpenML promotes reproducibility and collaboration across the research community, ensuring that scientific progress in machine learning remains open and accessible to all.

This seminar offers researchers a comprehensive and practical introduction to Automated Machine Learning (AutoML) and the OpenML platform, equipping participants with the tools and knowledge to integrate these methods into their own research projects.

The seminar opens with a plenary introduction to AutoML, covering its motivation, core techniques, and real-world applicability. Key methods such as TabPFN will be discussed, illustrating how AutoML can streamline machine learning workflows across a wide range of research domains.

The second session shifts to hands-on practice, walking participants through concrete code examples and guiding them in applying AutoML tools to their own datasets and research questions.

The seminar concludes with an exploration of advanced topics, including ongoing research projects within our group and open discussion of directions for future work, encouraging participants to identify opportunities to extend and build upon existing methods.

Saber Salehkaleybar is an Assistant Professor at Leiden University. Prior to joining Leiden University, he was a research scientist at EPFL.

Salehkaleybar’s research focuses on causal inference, stochastic optimization, and reinforcement learning. His work in causal inference is centered on two main themes: causal inference under latent confounding and experiment design in complex systems. He has developed methods for causal discovery and causal effect estimation when hidden variables may affect the observed data, and for designing efficient experiments in complex networks. Several of these contributions have appeared as spotlight papers at NeurIPS.

More broadly, his research aims to make causal reasoning reliable in realistic settings where standard assumptions may fail, and connects causal inference with modern AI and decision-making systems.

Causal inference provides tools for answering questions that go beyond prediction: What would happen if we intervened? Which relationships are truly causal rather than merely correlational? And how can we make reliable decisions when randomized experiments are impossible, costly, or unethical?

The seminar will begin with an introduction to the foundations of causal inference, including causal graphs, structural causal models, interventions, counterfactuals, causal discovery, and causal effect estimation. As these concepts are introduced, we will work through illustrative examples of causal discovery and causal effect identification, explaining key challenges that arise in real-world settings, such as latent confounding, biased observational data, and identifiability.

The seminar will conclude by exploring opportunities for causal inference in modern AI, with a particular focus on reinforcement learning and generative AI. We will discuss how causal ideas can support hierarchical reinforcement learning, mechanistic interpretability in generative AI, and more robust AI systems, while connecting these topics to recent research on causal discovery and causal effect estimation.

Yingjie Fan is an Assistant Professor at LIACS who leads the group of Logistics Optimization. Her research interests include stochastic programming, multi-objective optimization, robust optimization, learning-based optimization, behavioral operations research and data-driven modeling in transportation science and supply chain risk management.

Yingjie Fan is representing the Natural Computing Cluster for visiting Chile. At the Natural Computing Cluster (NaCo), part of LIACS at Leiden University, we delve into the fascinating intersection of nature-inspired computation and real-world problem-solving. Our cluster integrates six special interest groups, each pushing the boundaries of Artificial Intelligence, Machine Learning, Optimization, and Data Science.

Our mission is to foster a deep understanding of natural computing and to develop innovative solutions that can address some of the world's most complex problems. By harnessing the power of nature-inspired computing, we aim to contribute to scientific advancement and create a positive impact on society. Research in natural computing is at the core of our activities, covering theoretical foundations, the development of new algorithms, and interdisciplinary applications of natural computing methods. Our cluster focuses on a wide array of topics, combining foundational research with practical applications. These include:

• Nature-Inspired Algorithms: foundational work in evolutionary computing, metaheuristics and related methodologies.
• Explainable AI (XAI): Advancing fair, interpretable, and transparent AI solutions tailored to diverse industries.
• Algorithm Benchmarking and Configuration: Rigorous evaluation and fine-tuning for optimal algorithmic performance.
• Generative AI and LLM-Aided Design: Exploring creative and automated approaches for designing algorithms.
• Quantum Computing: Tackling optimization challenges using cutting-edge quantum methodologies.
• Logistics and Sustainability: Machine learning-driven solutions for supply chain, vehicle routing, and green manufacturing.
• Multi-Objective Optimization: Efficient techniques for robust, multimodal, and scalable optimization challenges.

The driving force behind our research is the mission to increase our understanding of natural systems as models of computation, with a focus on the development of new algorithms and applications to challenging problems.

In this seminar, I will present my research topics and those of the NaCo cluster, covering a broad range of areas, including multi-objective optimization, evolutionary computation, Bayesian optimization, metaheuristics, learning-based optimization, hyperparameter optimization, automated algorithm design using LLMs, explainable AI, and decision-support applications in engineering, medicine, supply chains, logistics, and hydrogen transport and production.

Alfons Laarman is an associate professor at Leiden University, where he leads the theory cluster and is co-founder of the inter-departmental Applied Quantum Algorithms (aQa) initiative. Prior to joining Leiden, he was a postdoc at TU Vienna and completed his PhD at the University of Twente.

Laarman is known for parallel DFS-based graph algorithms and for knowledge representation and reasoning approaches to quantum computing. In general, his research spans reliable computing, parallel computing, and quantum computing, often bridging theory and practice. His research is funded by several national and European consortia as well as by personal grants from the Dutch Scientific Research Council.

The first part of this lecture comprises a mini-tutorial on quantum computing. This tutorial is aimed at anyone with a computer science background and an interest in quantum computing. Starting with classical circuits as a computational model, we gradually introduce reversible and probabilistic computing as natural extensions for which computer scientists already have a strong intuition. We then show that probabilistic behavior breaks the reversibility required for quantum computing. The solution is to switch to a norm that allows for “negative probabilities”. With only two simple gates, we can then gain a full understanding of universal quantum computing.

In the final part, we will discuss various knowledge representation and reasoning techniques for analyzing and optimizing quantum circuits. Using model counting (#SAT), we show how to classically simulate quantum circuits and verify their equivalence. We then introduce a solution for quantum circuit synthesis based on maximum model counting. If time permits, we will relate the succinctness and tractability properties of tensor networks and decision diagrams, two formalisms often used to represent quantum information.

Dr. Akrati Saxena is an Assistant Professor at the LIACS, Leiden University, and an Adjunct Professor at the University of Victoria, Canada. Her research lies at the intersection of social network analysis, complex networks, computational social science, data science, and algorithmic fairness. She established and leads the AlFa (Algorithmic Fairness, https://alfa.liacs.nl/) research group, which focuses on understanding structural inequalities in complex networks and advancing fairness-aware algorithms in network and data science, including analyzing biases in existing systems, defining fairness constraints and evaluation metrics, and designing fair computational frameworks. She is actively building a research community on algorithmic fairness in network science and has organized several satellite events, thematic sessions, and tutorials to support this effort. She is an Associate Editor for SNAM and the PLOS Complex Systems journals. In addition to her research, she serves on the Diversity Committee at LIACS, contributing to initiatives that foster inclusion and equity within the academic community.

Complex networks, such as social, financial, e-commerce, and criminal networks, provide a powerful framework for representing real-world systems by capturing intricate structural patterns and interactions, consisting of nodes (entities) and edges (connections or interactions). For example, in social networks, nodes represent individuals, and edges denote social connections, while in banking transaction networks, nodes correspond to bank accounts, and edges represent financial transactions. Complex networks are analyzed to understand individual and group behavior at a large scale and solve critical research problems, such as fraud detection, link prediction, social media surveillance, and resource allocation. However, these networks often encode structural inequalities related to gender, ethnicity, race, or socioeconomic status. Moreover, groups’ distribution may be inherently imbalanced, with certain groups being underrepresented or more sparsely connected. If such structural inequalities are not considered while designing network analysis algorithms, the outcome might be unfair, particularly disadvantaging minorities or underrepresented groups.

In this lecture, I will highlight how the structural inequalities of complex networks impact the fairness of different network analysis methods using a case study of link prediction. Next, I will discuss a few approaches in depth to address structural biases for fair and diverse link prediction, specifically network-embedding-based methods. Finally, I will briefly introduce other approaches for developing fair solutions across diverse downstream network analysis tasks, along with the primary research focus of our group.

This final session is designed to facilitate discussion between participants from Leiden and Chile on technical topics, as well as on future opportunities for research collaboration, scholarships & academic/student exchanges.

• Provide an overview of research topics here in Chile.
• Identify shared research interests between Leiden and Chilean participants.
• Explore possible collaborations, joint projects, and follow-up activities.
• Identify opportunities for student/academic exchange.