Paolo Barucca, PhD

Associate Professor at University College London

[
[
[

]
]
]

Explainable Complexity Lab

My research combines statistical physics, network science, and machine learning, with a particular emphasis on understanding complex systems. I aim to develop principled, physics-inspired models that equip data-driven methods with robustness, interpretability, and theoretical foundations, especially when applied to highly interconnected and non-stationary environments.

My work is organized around three central themes:
(i) complex networks modeling (e.g., modeling temporal networks, contagion processes, liquidity cascades, and network evolution in financial markets),
(ii) statistical learning via random matrix theory and disordered systems (e.g. extracting signal from noise in high-dimensional data, characterizing spectral features of covariance structures), and
(iii) machine learning for dynamical systems (e.g. adapting learning under non-stationarity, combining physics-informed architectures with data-driven components, applying these to physical systems, forecasting and risk assessment).

Before joining UCL as Associate Professor, I obtained my PhD in Theoretical Physics from Sapienza University of Rome and developed my early research projects in the statistical physics of disordered systems. Over time, my focus has shifted toward the intersection of physics, network theory, and machine learning, applied especially to economics and finance. I have collaborated with scholars across disciplines to tackle questions of stability, inference, and predictability in complex systems.

Interests
Complex Networks & Systemic Risk
Statistical Physics & Random Matrix Theory
Physics-informed Machine Learning
Dynamics and Non-stationarity
Robustness, Interpretability, and Generalization

Education & Academic Position
PhD in Theoretical Physics, Sapienza University of Rome
Associate Professor, Computer Science Department, University College London

Team

PhD Students

Kentaro Hoshisashi is a PhD student at UCL working at the intersection of machine learning, quantitative finance, and physics-informed modeling.
He introduced Whack-a-mole Online Learning (WamOL), a physics-informed approach for real-time implied volatility surface calibration. His work enforces no-arbitrage and PDE constraints within deep learning frameworks for option pricing.
Overall, his research aims to build robust, theory-consistent ML tools for high-frequency financial modeling.

Marcelina Marjankowska is a PhD student in the EIGENDATA project, focuses on the theoretical and computational development of eigenvector statistics in random matrix ensembles and their application to Physics-Informed Neural Networks. Her research contributes to deriving finite-size analytical results, implementing numerical simulations, and validating new spectral methods in statistical inference and optimization.

Teaching

Workshops and Seminars

Engagement and Outreach

Books

News

Contact

p dot barucca at ucl dot ac dot uk