"Towards secure Artificial Intelligence: Private distributed learning and strategic decision making"

Institute of Information Theory and Automation (UTIA), The Czech Academy of Sciences, Seminar

Prague, Czech Republic — November 2025

Artificial intelligence faces security challenges at many levels, such as the exposure of sensitive data, the vulnerability of distributed learning systems, and the need to design robust policies under adversarial uncertainty. In this seminar, I discussed two approaches to improve AI security: federated learning to preserve privacy across distributed data sources, and Adversarial Risk Analysis (ARA) to formulate decision-theoretic defensive planning against stochastic adversarial agents.

"Tracking Latent Goals for Robust Reinforcement Learning"

II International Workshop on Bayesian Statistics (SEIO Bayesian Working Group) — Contributed Session IV: Adversarial Learning & Robust ML

CUNEF Universidad, Madrid, Spain — June 4–5, 2026

Talk presented in the session on Adversarial Learning & Robust ML, exploring methodology for tracking latent goals to improve policy robustness in reinforcement learning under strategic uncertainty.

"Opponent-Aware Soft Q-Learning"

NextGen Synergy: Control Theory & Machine Learning (COML 2026 / InterCoML) — COST Action CA24136

Prague, Czech Republic — April 27–30, 2026

Oral presentation on Opponent-Aware Soft Q-Learning with David Ríos Insua. Additionally participated as an invited discussant in Roundtable I: Adversarial, Risk-Aware Decision Making in Multi-Agent Systems (moderated by David Ríos Insua, alongside Leon Bungert and Agnieszka Wiszniewska-Matyszkiel).

"Adversarial Risk Analysis with Fully Probabilistic Designs to handle sequential games"

EURO 2025 Congress

Leeds, United Kingdom — June 2025

Adversarial risk analysis (ARA) is a decision analytic methodology that informs decision-making when facing intelligent opponents and uncertain outcomes. It enables an analyst to model beliefs about an opponent's utilities, capabilities, probabilities, and strategic calculations. In this presentation we discussed how to merge ARA and Fully Probabilistic Design to handle sequential games.

"Adversarial Risk Analysis with Fully Probabilistic Designs"

XLI National Congress of Statistics and Operations Research (SEIO 2025)

Lleida, Spain — June 2025

ARA is applied to defend-attack games mitigating standard common knowledge and common prior assumptions in classical game settings. Yet it entails complicated modeling and computations. In this work we discuss a new approach to ARA using a Fully Probabilistic Design methodology in simple defend-attack games.

"A new federated learning adaptation of AdaBoost"

AEMCO Congress

Sevilla, Spain — September 2024

Federated Learning allows training artificial intelligence models with distributed data, respecting privacy and accessing large volumes of information without exposing sensitive data. This conference talk introduces Federated Learning (FL), gives an example of a novel FL algorithm and highlights possible FL applications to Health and Social Sciences.

"Delayed Spiking Neural Networks for Neuromorphic Computing"

AIHUB's Connexion Summer School

Valencia, Spain — July 2024

Spiking Neural Networks (SNNs) offer a biologically plausible alternative to ANNs by computing in discrete spikes instead of continuous activations. Incorporating learnable synaptic delays enhances their temporal learning while reducing required parameters. This poster covers an empirical study on the performance of delay-augmented SNNs on neuromorphic spatio-temporal datasets.

"Rethinking the Role of Bayesianism in the Age of Modern AI"

International Centre for Mathematical Sciences (ICMS)

Edinburgh, United Kingdom — October 2026

Participation in the workshop focused on Bayesian principles, foundational challenges, and scalable probabilistic inference in modern artificial intelligence systems.

Reinforcement Learning & Stochastic Control Programmes

Isaac Newton Institute for Mathematical Sciences

Cambridge, United Kingdom

• Workshop on Reinforcement Learning for Science: Discovery and Automation (2026)
• Bridging Stochastic Control and Reinforcement Learning: Theories and Applications (2025)

Reinforcement Learning & MARL Reading Group

Community Reading Group — Mathematical Foundations & Frontiers of Decision Making

A reading group focused on Reinforcement Learning (RL), Multi-Agent RL (MARL), and sequential decision-making. We follow Kevin P. Murphy's Reinforcement Learning: An Overview along with recent research papers across value-based methods, policy gradients, model-based RL, and multi-agent game-theoretic frameworks.

Probabilistic Machine Learning Reading Group

Community Reading Group — Bayesian Modeling & Inference

Bi-weekly collaborative reading group exploring Probabilistic Machine Learning: Advanced Topics by Kevin P. Murphy. Covered advanced Bayesian methods, approximate inference, variational techniques, and modern probabilistic deep learning with open materials and discussions.