Geometry & vision
Spectral methods, deformable shape analysis and learned representations of 3D data.
Computer science · Ca’ Foscari, Venice
Luca Cosmo
Associate Professor of Computer Science
Department of Environmental Sciences, Informatics and Statistics (DAIS)
I study how geometry and relational structure can improve machine learning. My research spans computer vision, geometric deep learning and graph learning, with applications to 3D shapes, graphs, music and scientific data.
Research
All publications →Graph learning
Interpretable graph convolutions, graph kernels and structural attention for relational data.
Generative models
Diffusion models for graphs, musical audio, molecular design and 3D reconstruction.
Selected research
Representative work and directions for further research
Graph generation ICLR 2025
Generating Graphs via Spectral Diffusion
Generating graph structures by learning to sample Laplacian eigenvalues and eigenvectors.
Interpretable graph convolutions IEEE TNNLS 2025
Graph Kernel Neural Networks
Graph kernels extend convolution to irregular graphs. The learned structural masks offer a way to inspect which graph patterns the network responds to.
Background
I have been an Associate Professor at Ca’ Foscari since October 2024, following three years as a tenure-track researcher.
Previously, I was a postdoctoral researcher with Emanuele Rodolà at Sapienza University of Rome, working on shapes and graphs within the ERC project SPECGEO. In 2019, I worked in Michael Bronstein’s Geometric Deep Learning group in Lugano. Earlier, I was a postdoctoral researcher in Ca’ Foscari’s Computer Vision group.
I received my PhD in Computer Science from Ca’ Foscari. During my PhD, I visited Daniel Cremers’ group at TU Munich, where I began working on deformable shape analysis and spectral methods.
Contact
Get in touch
For research collaborations, course questions or thesis enquiries.
Scientific Campus, Via Torino 155
30172 Venezia Mestre, Italy
Office Z.B08, Zeta B building
Office hours By appointment