Previous work has shown that using web images within the framework of Current Weakly-Supervised Incremental Learning for Semantic Segmentation (WILSS) can effectively retrieve images of previously seen classes and learn new classes, achieving state-of-the-art performance.
In particular, we propose an architecture that relies on vision-language models to extract captions from images, which are useful for searching the web for images of both old and new classes. This thesis aims to develop new ...
In recent years, Neural Radiance Fields (NeRF) have been used to reconstruct a 3D scene starting from RGB images. They represent a density function and a color function using neural networks which are trained by minimizing a reconstruction loss with respect to the input views. A limitation of NeRFs is that they require many input views to achieve a good reconstruction. To overcome this problem, many research works use generative models to optimize the NeRF parameters. These approaches often i...
This study investigates the impact of new bike lanes on urban traffic patterns on two specific streets in Cambridge (completed in 2017) and Boston (completed in 2018), Massachusetts. By examining data from the non-winter period during the morning and afternoon rush before and after the completion of the bike lane, the research analyzes various aspects of bike traffic to determine the changes brought by the new infrastructure. This study aims to determine if there is any difference in connecti...
In semiconductor manufacturing, spatial pattern recognition is essential for identifying defects or obtaining other crucial information on electrical wafer maps. During the wafer testing stage, deep learning methods are widely used for their powerful feature extraction capabilities.
The aim of this thesis is to propose a lightweight CNN model that achieves comparable or superior results to more complex models, enabling faster training and greater flexibility for experimentation and futurue i...
The candidate will consider the problem of efficiently decomposing a tripartite tensor into a small tensor network (3-6), with three tensors and six total links (three external, three internal). In the information perspective, an efficient decomposition is such that each correlation between the three parties takes the shortest route along the (3-6) network. The candidate will develop numerical algorithms to acquire such efficient tripartite decomposition, which in turn will play the role of a...
Metabolic dysfunction–associated steatotic liver disease (MASLD), previously known as non alcoholic fatty liver disease (NAFLD), is a chronic liver disease characterized by excessive fat accumulation in the hepatocytes, leading to liver steatosis and potential progression to more severe liver conditions, annually responsible of 1 out of every 25 deaths worldwide. Due to the lack of pharmacological and targeted treatments, this study aims to build a liver numerical digital twin, reproducing th...
This thesis aims to develop an innovative methodology for analyzing and representing the tactical profiles of Serie A football players using detailed match-by-match statistical data. The primary objective is to delineate player behavior through the generation of various types of heatmaps that capture different types of actions. These heatmaps will be reduced in dimensionality, projecting playing styles into vector form, using two different approaches: a Variational Autoencoder and a Non-Negat...
The human microbiome, a complex and dynamic ecosystem of microorganisms, plays a crucial role in health and disease. Understanding its dynamics is essential for developing targeted therapies and diagnostic tools. This thesis investigates the universality of human microbiome dynamics, with a particular emphasis on Dissimilarity Overlap Analysis (DOA), a novel analytical approach designed to elucidate the underlying patterns and transitions within microbial communities. By applying DOA, we quan...
With the advent of the current era of Noisy Intermediate-Scale Quantum (NISQ) devices, new computational paradigms are being explored to leverage the developing quantum technologies. Hybrid quantum-classical algorithms, such as the variational quantum eigensolver or the variational quantum simulation, have emerged as a promising tool, for example, to accurately determine properties of molecules on a quantum computer or a quantum simulator, respectively. This is done by combining a quantum dev...
This thesis presents an analysis of the energy resolution in the T2K (Tokai to Kamioka) experiment, in particular at the Super-Kamiokande (SK). The T2K experiment, designed to study neutrino oscillations, employs the SK detector to measure neutrino interactions. Accurate energy reconstruction is crucial for understanding these interactions and improving the precision of oscillation parameter measurements. By optimizing energy resolution, this study aims to enhance the accuracy of neutrino ene...
