I am interested in Graph Neural Networks: with my collaborators, we are investigating their applications to time series. We are also working on applications of Sheaf Theory and Geometry to Machine Learning and AI. We combine these techniques and theoretical ideas with Deep Learning and other standard Machine learning techniques (regression, SVM, clustering algos, random forests, etc.).

Projects I am working on:

Deep Learning for ophthalmic data analysis

• Collaboration with Studio Oculistico D’Azeglio: Developed DL algorithms for hands-on clinical deployment, focusing on the diagnosis and monitoring of rare disease progression. Two abstracts submitted on the subject to the European Neuro-Ophthalmology Society (EUNOS) conference, June 2026.

• Autoencoders for visual fields reconstruction, with archetypal analysis on the latent space.

• Transfer learning and development of DL algorithms in data-scarce regimes (rare diseases).

Data-Aware Geometric generative deep learning methods 

• Sheaf Neural Networks: Developing geometric frameworks using sheaf theory to model complex interactions and heterophilic relations in graph-structured data.

• Generative Modeling: Researching geometry-based Flow Matching techniques to optimize continuous-time normalizing flows, improving sampling efficiency and training stability. Integrating Compositional Data Analysis (CoDA) to enforce simplex constraints and enhance model interpretability.

FRAIL project: medical data and imaging 

Funded by the EU grant FRAIL - Fracture Risk evaluation in bone metastatic patients by Artificial InteLligence

• Main developer for the project FRAIL, aimed at prediction of survival of cancer patients with bone metastases 1 year after surgery and fracture risk assessment in metastatic long bones.

• Data engineer: organized and formatted the raw data provided by IOR for its effective usage in Deep Learning and Machine Learning. Handled both clinical data in tabular form and medical images (both x-rays and CT scans).

Geometric Deep Learning: anomaly detection in biological signals

• End to end deployment of an anomaly detection algorithm for biological signals (ECG and PPG) for a wearable health-monitoring device. In partnership with with IppocraTech - VST.

• Advanced the theory of the mathematical foundations of Geometric Deep Learning (paper submitted for review). Joint with University of Bologna.

• GNN-Enhanced TCN Algorithms for ECG Signal Quality Recognition, Geometric Science of Information Conference (GSI), 2025

• Kolmogorov-Arnold Convolutional Network models for supervised quality recognition of PPG signals, A. Mehrab, M. Lapenna, F. Zanchetta, A. Simonetti, G. Faglioni, A. Malagoli, R. Fioresi, Entropy, 2025

• Graph Neural Networks and Time Series as Directed Graphs for Quality Recognition, A. Simonetti, F. Zanchetta. Preprint,  arXiv:2310.02774, 2023

• A Geometric deep learning approach to blood pressure regression, F. Zanchetta, A. Simonetti, G. Faglioni, A. Malagoli and R. Fioresi. Extended abstract accepted to GeoMedIA workshop 2022. Link 

• GNN-Enhanced TCN Algorithms for ECG Signal Quality Recognition, GSI 2025, October 2025.

• ˆ Machine learning for biomedical data, Modena AI, UNIMORE, April 2025.