Doctoral Candidate

Leonardo Schiavo,  MSc

Leonardo Schiavo received his Master’s degree in Data Science from the University of Padua. He is a sporty and sociable Data Scientist–Mathematician with a passion for theory and research. He is always seeking challenges that allow him to apply his skills in creative problem-solving. He is currently a Marie Curie PhD student at the Technical University of Madrid (UPM), working on XAI and time series.

Leonardo Schiavo_Self-presentaion.pdf

Main Supervisors: David Camacho (UPM), Ángel Panizo Lledot (UPM)

Co-Supervisors: Salvatore Cuomo (UNINA), Dariusz Mrozek (SUT), Volker Stolz (HVL)

R&D cooperation: GMV

Objectives: advance anomaly detection in complex multivariate temporal data by combining deep generative models and explainable AI (XAI) principles. 

Many real-world domains such as industrial monitoring, healthcare, and finance rely on high-dimensional time series where anomalies arise from intricate inter-variable dependencies. This research focuses on developing transparent, scalable, and interpretable models that can both detect and explain anomalies across correlated variables.

The work extends previous univariate anomaly detection frameworks by introducing multivariate adaptations and state-space architectures. In particular, the research explores the integration of Mamba, a recent state-space deep learning architecture, underutilized in the literature. Mamba can be integrated within a masked latent generative modeling framework to improve both speed and memory efficiency in long and high-dimensional sequences. Explainability will be a key component, allowing the identification variables and temporal regions contribution to each detected anomaly.

Expected Results: 

A novel, explainable multivariate anomaly detection framework combining generative modeling and state-space architectures.

Advanced evaluation methodology for anomaly interpretability and cross-variable dependencies.

Publications in leading AI and data mining venues (e.g., ICML, ICLR, Information Fusion).

Contributions to TUAI’s WP2 objectives, enhancing explainable and trustworthy AI for time series applications.

Applied research: The primary outcome will be new deep learning-based time series data techniques, the project will be applied to  industrial sectors (e.g. space under the collaboration with GMV), to validate the practicality and feasibility of the new time series models.

Planned secondments: UNINA (4 months); SUT(4 months);HVL(4 months)                    

Enrolment in Doctoral degree: UPM