Hierarchical Classification in Low-Data Settings

Machine Learning has made significant strides in various domains, with classification tasks playing a crucial role in applications such as sentiment analysis, image recognition, and spam detection. However, deep learning models often require large amounts of labeled data, which can be difficult to obtain due to cost, time, and privacy constraints. This challenge has led to the development of Low-Shot Learning techniques, including Few-Shot and Zero-Shot Learning, which enable models to genera...

ReGenT-seq identifies a chromatin-to- cell protrusion mRNA transport for epidermal cell activation

In response to stress, cells exploit plasticity to self-activate and form protrusions to explore the surroundings and guide migration. While the initiation of protrusions is well-characterized, little is known about their maturation and functional composition. Here, using a new pooled genetic approach (ReGenT-seq) to screen all the chromatin factors controlling epidermal stem cell activation, we unexpectedly identified Cbx3 as a critical protrusion founder. Beyond its nuclear activities, we r...

Co-occurrence of facial gestures and vocalisations across primates

Multimodal communication, intended as the sequential or simultaneous integration of signals across different communicative modalities, is a key aspect of primate social behaviour. Among these, the interplay between facial gestures and vocal emissions has long been hypothesised to have evolutionary relevance for the emergence of complex communication systems, including human language. Nevertheless, empirical studies investigating the co-occurrence between different modalities remain scarce, in...

Green analysis for quality fragrances: strategies for the improvement of the sustainability of the analytical process in the fragrance field from sample preparation to analysis

Sviluppo della missione spaziale NUSES per l’osservazione in UV della Terra e per la rivelazione di impulsi elettrici di origine pre-sismica

INNOVATIVE EXPERT-BASED TOOLS FOR MULTITEMPORAL GEOENVIRONMENTAL ANALYSES: FROM MAPPING TO MODELLING

Accurate, timely, and scalable geomorphological risk assessment strategies are essential for effective territorial management and mitigation planning. This dissertation addresses the limitations of traditional geomorphological mapping, especially regarding landslide risk assessment by developing and integrating novel, open, and interoperable tools that blend expert knowledge with artificial intelligence. A human-centred approach guided this doctorate project, where automated technologies supp...

Biotechnological approaches for genetic improvement in Solanaceae species

Solanaceae is one of the most valuable and variable families among vegetable crops and includes some of the most important agricultural species (potatoes, tomatoes, eggplants and peppers) and some crop model plants for research (e.g. tomato and tobacco). In order to meet the changing and increasing needs of agriculture, new biotechnological tools can be exploited to support breeding programs and to facilitate the development of new crop varieties with improved traits. The purpose of this thes...

Disturbance Footprint in Fragmented Tropical Rainforests: An Integrated Approach to Ecosystem Health Assessment in Madagascar

APPLICATION OF REMOTE SENSING FOR THE YIELD AND QUALITATIVE IMPROVEMENT OF CEREAL CULTIVATION FOR ADVANCED SUPPLY CHAINS

This PhD thesis investigates the application of remote sensing (RS) and proximal sensing technologies for improving nitrogen (N) management in wheat and maize cultivation. The general objective is to optimize yield and grain quality while reducing environmental impact, supporting sustainable precision agriculture (PA) practices. Chapter 1 introduces the potential of RS in agriculture, explaining how vegetation indices (VIs) derived from both proximal and satellite sensors can support N manage...