work with me

Thesis and research opportunities

This is a list of topics I am interested in. The list is not exhaustive! If you are interested in a particular topic, email me to discuss a potential research project or BSc/MSc/PhD thesis.

Topics

  • 3D Computer Vision for pointcloud analysis using Graphs and Transformers
    We design synthetic dataset generation pipelines using Blender python API in headless mode, customising the simulation of erosion via PDEs. We use Graph Neural Networks, PointNet++, and Transformer models to perform multiple (parallel) tasks on pointclouds/meshes: classification, orientation regression, fragment pairings. In collaboration with INFN, Torino, and UniFi.

  • Generative models for MRI to synthetic CT scans
    We design multiple Pix2Pix-like architectures for Magnetic Resonance Imaging to synthetic Computed Tomography scans, to assist radiologist in their definition of the radiotherapy. We focus on stabilising the adversarial training, use additional multi-objective loss, especially perceptual, and, last but not least, architectural changes like the usages of Adaptive Fourier Neural Operators layers. We are currently working in an attention-free transformer-like UNet architecture for the backbone.

  • Computer Vision for Spectral Data
    We study how to apply Computer Vision deep learning methods on Spectral datacubes analysis with focus on Cultural Heritage and Astrophysics. We work with Pix2Pix approach with AFNO-based and/or ViT-based backbones. A great effort is dealt on understanding a possible Latent space representation of spectral data, using techniques such as Deep Variational Embedding. We are exploring the usage of domain adaptation techniques (either algorithmic, such as Fourier Domain Adaptation, or unsupervised/adversarial, as DANNs) to reduce the gap betweem syntehtic data and real data.

  • Physics Informed Neural Networks & Neural Operators
    We work on PINNs and NOs design for data-driven/pure-forward resolution of various PDE systems, from 3D diamond detectors design to Dirac-Born-Infeld systems for QCD phase transition. We use PyTorch, Lightning, and NVIDIA PhysicsNeMo. In collaboration with ENI.

  • Deep Learning for Genetic Data
    We would like to apply ML techniques to genetica data to infer some properties, e.g., the biogeographical ancestry. This project is financed via a FIS3 funding obtained by Prof. Elena Pilli. In collaboration with UniFi.

  • Design of efficient, cloud native web application for physics data analysis
    We design cloud-native applications using Dash/Fast API/Jax.

the usage of plural is because all the activities are performed within a team. Not always is the same team; usually, each topic is a different team/collaboration.