About Me

Note: This website is currently under construction. Content and features may change as the site is updated.

Hi, I’m Luca — a PhD student in Computer Science at the University of Pisa, working with Prof. Davide Bacciu. I spend my time somewhere between deep generative models and graph machine learning.

Lately I’ve been interested in how you can compose and steer generative models without retraining them. In CompFlow (CompLearn @ ICML 2026) we show that simply adding up conditional velocity fields at inference time turns out to be a tempered Product-of-Experts — which takes classifier-free guidance from one condition to as many as you like, for free. It even works zero-shot on top of large text-to-image models. Before that, I used energy-based models on molecular graphs to optimise chemical properties implicitly, with no explicit optimisation loop.

On the graph side, I’m curious about where message passing breaks down. In Can You Hear Me Now? we built a benchmark that measures how far information actually travels in a graph neural network, instead of guessing it from downstream accuracy.

Before the PhD I did my Master’s here in Pisa (110/110 cum laude), and worked as a research fellow on AI for healthcare. That meant a lot of smartphone accelerometers: measuring respiratory kinematics and spotting cardiorespiratory conditions with deep learning, published in Electronics and Computers in Biology and Medicine and funded by Regione Toscana.

Location

You can find me around here:

Dipartimento di Informatica, Università di Pisa
Largo Bruno Pontecorvo, 3, 56127 Pisa PI, Italy