Mattia Matucci
Nationality: Italian
Host: Inria Université Côte d’Azur
| Background: After completing a Bachelor’s degree in Computer Science at the University of Florence, he continued his studies by enrolling in a Master’s degree in Computer Science with a Data Science curriculum at same university. During the Master he studied a broad range of subjects, including computer aided geometric design, machine learning, parallel programming and advanced programming techniques and tools. He recently graduated with a thesis focused on the study and development of generative neural networks for point clouds. Post-graduation, Mattia delved deeper into the topics covered in the thesis. He also spent a couple of weeks at the TU Delft on an Erasmus program, where he investigated an alternative type of generative neural networks for point clouds. |
| Master thesis: Generative Adversarial Networks for Point Cloud Generation Abstract: This thesis explores a solution to the limited data availability for point cloud processing in geometric design applications, through the use of Generative Adversarial Networks (GAN). It presents a detailed analysis of selected models, including their validation on the ShapeNet dataset, followed by their fine-tuning for application to point clouds sampled from rectangular surface patches. In addition, a novel method for extracting boundary parametrization from these patch-like point clouds is proposed. The proposed point cloud generation model is also validated through the application of the Boundary-Informed Dynamic Graph Convolutional Network (BIDGCN) to the problem of parameterizing three-dimensional unstructured data sets over a planar domain. The BIDGCN network, trained on the newly generated point clouds, achieves comparable performance with respect to training on synthetic data, proving that GANs can effectively be used for Data Augmentation also for point cloud processing applications. |
| Research interests: Computer Aided Geometric Design, particularly the development of algorithms and data structures for geometric processing and adaptive spline spaces, and Machine Learning, with a primary focus on geometric deep learning and generative models. |
| Goals in TENORS: His PhD project focuses on overcoming the curse of dimensionality of high-dimensional PDEs. He aims to achieve this by developing innovative algorithms and data structures, while also leveraging his studies in machine learning. As part of the TENORS network, he would like to foster a collaborative research environment where sharing different backgrounds and interests allows everyone to explore their topics with new perspectives. |
| Hobby: sports, especially going to the gym, spending time with my friends, watching movies and playing video games. |
