Shape matching is a core task in many fields such as computer graphics, robotics, medical imaging, and geometric learning. Accurately establishing correspondences between 3D shapes enables applications ranging from surgical planning to robotic grasping and manipulation. In this talk, I present a data-driven perspective where understanding geometry and learning shape correspondences become mutually reinforcing goals. By targeting tasks such as shape matching, deformation, and representation refinement, we show that machine learning captures geometric features, while at the same time, geometry itself guides learning towards robust and interpretable models.

We explicitly exploit geometric structures within data-driven pipelines, showing that localized deformation cues can be learned without relying on global shape encodings. This integration of geometric priors into neural architectures leads to more efficient and interpretable systems suited for real-world applications at scale.

Going further, we explore the emergent structure within learned representations and attention patterns in Transformer-based models for point cloud matching. These insights enable us to reduce training overhead and improve robustness by integrating geometry-aware mechanisms directly into the model design. Our findings pave the way for future research that unifies geometric priors and deep learning to build AI systems that not only learn from shape but understand it, enabling smarter tools in design, medicine, and robotics.

Short Bio: 

Simone Melzi is an Associate Professor at the University of Milano-Bicocca, in the Department of Informatics, Systems and Communication (DISCo). He was a Post Doctoral researcher in Computer Science at the Sapienza University of Rome in the GLADIA group led by Emanuele Rodolà, at the École Polytechnique in the team of Prof. Maks Ovsjanikov (6 months) and at the Università Degli Studi di Verona (2018-2019). He received his PhD in Computer Science at Università Degli Studi di Verona (2018) and graduated in math at the University of Milan "La Statale" (2013). He received The Marie-Curie Individual Fellowships H2020-MSCA-IF-EF-ST-2019, the BE-FOR-ERC 2020 grant from the Sapienza University of Rome, the EG-Italy PhD thesis Award 2018 and the Eurographics Young Researcher Award 2023. He is a member of the Junior Fellow of Eurographics and a scholar of the European Lab for Learning and Intelligent Systems (ELLIS). He works on geometry processing, 3D shape analysis and artificial intelligence.

-- 
---------------------------------------------------------------------
Associate Professor, Loris Roveda

email: loris.roveda@supsi.ch
webpage:
https://orcid.org/0000-0002-4427-536X
https://scholar.google.it/citations?user=3un_pPgAAAAJ&hl=it&oi=ao
https://www.scopus.com/authid/detail.uri?authorId=56031943800

Istituto Dalle Molle di studi sull’Intelligenza Artificiale (IDSIA)
Scuola Universitaria Professionale della Svizzera Italiana (SUPSI)
Università della Svizzera italiana (USI)

Politecnico di Milano, Mechanical Department
---------------------------------------------------------------------