I am a PhD student in Smart Computing and Artificial Intelligence at the University of Florence, working in collaboration with the University of Siena. My current research focus is fake and synthetic-content detection, especially for images: I study which representations preserve useful traces of the generation process and how those traces behave across models, datasets, and common image transformations.
Research directions
One part of this work looks inside generative models rather than treating them only as black-box image generators. Intermediate features, attention patterns, and diffusion timesteps can expose information that is lost at the final output. I am interested in turning these observations into detection methods that remain useful outside a single, carefully controlled dataset.
Alongside this main direction, I work on machine-learning applications in biomedical and biological settings, particularly problems involving images, molecules, and structured scientific data. I am also interested in graph neural networks and graph learning as tools for representing relations that are difficult to express with fixed-size feature vectors.
These directions overlap in practical ways. Detection research raises questions about learned representations and their reliability. Biomedical work requires careful modelling and evaluation on complex, often limited data. Graph methods provide a natural language for molecules, biological systems, and other domains where relations are part of the signal rather than additional metadata.
Academic setting
My work is connected to the Siena Artificial Intelligence Lab (SAILab) at the University of Siena, where I am advised by Franco Scarselli and Monica Bianchini. SAILab has longstanding expertise in graph learning, neural networks for structured data, bioinformatics, and biomedical image analysis. This makes it a useful setting for research that moves between methodological questions and applications without losing sight of either.
I try to keep each project grounded in a clear experimental question: what information the model receives, what it learns, how it is evaluated, and which conclusions the evidence actually supports. Reproducible code and explicit limitations matter to me because both detection and scientific machine learning are particularly sensitive to shortcuts in the data.
Background
Before the PhD, I studied Artificial Intelligence and Automation Engineering and Computer Engineering at the University of Siena. Those studies led me from software and network projects to computer vision, probabilistic modelling, and research on structured data. They also taught me to value small, working tools alongside larger research projects.
The projects collected here follow that path. Some are published research, others are software or course experiments, but together they show how my current work connects representation learning, fake detection, structured data, and biomedical applications.