Intermediate linguistic task fine-tuning on Multi-lingual models

State-of-the-art NLP systems are generally based on the assumption that the underlying models are provided with vast datasets to train on. However, especially when working in multi-lingual contexts, datasets are often scarce, thus more research should be carried out in this field. This thesis investigates the benefits of introducing an additional training step when fine-tuning NLP models, named Intermediate Training, which could be exploited to augment the data used for the training phase. The Intermediate Training step is applied by training models on NLP tasks that are not strictly related to the target task, aiming to verify if the models are able to leverage the learned knowledge of such tasks. Furthermore, in order to better analyze the synergies between different categories of NLP tasks, experimentations have been extended also to Multi-Task Training, in which the model is trained on multiple tasks at the same time.

Spectral distortions of the cosmic microwave background and small-scale gravitational waves as new windows on the Early Universe

In this thesis, we explore constraints which can be put on the primordial power spectrum of curvature perturbations beyond the scales probed by anisotropies of the cosmic microwave background and galaxy surveys. We exploit present and future measurements of CMB spectral distortions, and their synergy with CMB anisotropies, as well existing and future upper limits on the stochastic background of gravitational waves. We derive for the first time phenomenological templates that fit small-scale bumps in the primordial power spectrum generated in multi-field models of inflation. By using such templates, we study for the first time imprints of primordial peaks on anisotropies and spectral distortions of the cosmic microwave background and we investigate their contribution to the stochastic background of gravitational waves. Through a Monte Carlo Markov Chain analysis we infer for the first time the constraints on the amplitude, the width and the location of such bumps using Planck and FIRAS data. We also forecast how a future spectrometer like PIXIE could improve FIRAS boundaries. The results derived in this thesis have implications for the possibility of primordial black holes from inflation.

New integer programming models for balanced multi-country kep

Le persone che soffrono di insufficienza renale terminale hanno due possibili trattamenti da affrontare: la dialisi oppure il trapianto di organo. Nel caso volessero seguire la seconda strada, oltre che essere inseriti nella lista d'attesa dei donatori deceduti, possono trovare una persona, come il coniuge, un parente o un amico, disposta a donare il proprio rene. Tuttavia, non sempre il trapianto è fattibile: donatore e ricevente possono, infatti, presentare delle incompatibilità a livello di gruppo sanguigno o di tessuto organico. Come risposta a questo tipo di problema nasce il KEP (Kidney Exchange Program), un programma, ampiamente avviato in diverse realtà europee e mondiali, che si occupa di raggruppare in un unico insieme le coppie donatore/ricevente in questa stessa situazione al fine di operare e massimizzare scambi incrociati di reni fra coppie compatibili. Questa tesi approffondisce tale questione andando a valutare la possibilità di unire in un unico insieme internazionale coppie donatore/ricevente provenienti da più paesi. Lo scopo, naturalmente, è quello di poter ottenere un numero sempre maggiore di trapianti effettuati. Lo studio affronta dal punto di vista matematico problematiche legate a tale collaborazione: i paesi che eventualmente accettassero di partecipare a un simile programma, infatti, devono avere la garanzia non solo di ricavarne un vantaggio, ma anche che tale vantaggio sia equamente distribuito fra tutti i paesi partecipanti.

Exploring the Higgs-muon coupling at a multi-TeV muon collider

We investigate the potential of a high-energy muon collider in measuring the muon Yukawa coupling (y_μ) in the production of two, three and four heavy bosons via muon-antimuon annihilations. We study the sensitivity of these processes to deviations of y_μ from the Standard Model prediction, parametrized by an effective dimension-6 operator in the Standard Model Effective Field Theory (SMEFT) framework. We also consider the κ framework, in which the deviation is simply parametrized by a strength modification of the μ+μ−h vertex alone. Both frameworks lead to an energy enhancement of the cross sections with one or more vector bosons, although the κ framework yields stronger effects, especially for the production of four bosons. On the contrary, for purely-Higgs final states the cross section is suppressed in the κ framework, while it is extremely sensitive to deviations in the SMEFT. We show that the triple-Higgs production is the most sensitive process to spot new physics effects on y_μ.

Zoomed simulations of Halo segregation in cosmological models with two species of coupled dark matter

La materia ordinaria copre soli pochi punti percentuali della massa-energia totale dell'Universo, che è invece largamente dominata da componenti “oscure”. Il modello standard usato per descriverle è il modello LambdaCDM. Nonostante esso sembri consistente con la maggior parte dei dati attualmente disponibili, presenta alcuni problemi fondamentali che ad oggi restano irrisolti, lasciando spazio per lo studio di modelli cosmologici alternativi. Questa Tesi mira a studiare un modello proposto recentemente, chiamato “Multi-coupled Dark Energy” (McDE), che presenta interazioni modificate rispetto al modello LambdaCDM. In particolare, la Materia Oscura è composta da due diversi tipi di particelle con accoppiamento opposto rispetto ad un campo scalare responsabile dell'Energia Oscura. L'evoluzione del background e delle perturbazioni lineari risultano essere indistinguibili da quelle del modello LambdaCDM. In questa Tesi viene presentata per la prima volta una serie di simulazioni numeriche “zoomed”. Esse presentano diverse regioni con risoluzione differente, centrate su un singolo ammasso di interesse, che permettono di studiare in dettaglio una singola struttura senza aumentare eccessivamente il tempo di calcolo necessario. Un codice chiamato ZInCo, da me appositamente sviluppato per questa Tesi, viene anch'esso presentato per la prima volta. Il codice produce condizioni iniziali adatte a simulazioni cosmologiche, con differenti regioni di risoluzione, indipendenti dal modello cosmologico scelto e che preservano tutte le caratteristiche dello spettro di potenza imposto su di esse. Il codice ZInCo è stato usato per produrre condizioni iniziali per una serie di simulazioni numeriche del modello McDE, le quali per la prima volta mostrano, grazie all'alta risoluzione raggiunta, che l'effetto di segregazione degli ammassi avviene significativamente prima di quanto stimato in precedenza. Inoltre, i profili radiale di densità ottenuti mostrano un appiattimento centrale nelle fasi iniziali della segregazione. Quest'ultimo effetto potrebbe aiutare a risolvere il problema “cusp-core” del modello LambdaCDM e porre limiti ai valori dell'accoppiamento possibili.

Modelling and simulating in systems biology: an approach based on multi-agent systems

Systems Biology is an innovative way of doing biology recently raised in bio-informatics contexts, characterised by the study of biological systems as complex systems with a strong focus on the system level and on the interaction dimension. In other words, the objective is to understand biological systems as a whole, putting on the foreground not only the study of the individual parts as standalone parts, but also of their interaction and of the global properties that emerge at the system level by means of the interaction among the parts. This thesis focuses on the adoption of multi-agent systems (MAS) as a suitable paradigm for Systems Biology, for developing models and simulation of complex biological systems. Multi-agent system have been recently introduced in informatics context as a suitabe paradigm for modelling and engineering complex systems. Roughly speaking, a MAS can be conceived as a set of autonomous and interacting entities, called agents, situated in some kind of nvironment, where they fruitfully interact and coordinate so as to obtain a coherent global system behaviour. The claim of this work is that the general properties of MAS make them an effective approach for modelling and building simulations of complex biological systems, following the methodological principles identified by Systems Biology. In particular, the thesis focuses on cell populations as biological systems. In order to support the claim, the thesis introduces and describes (i) a MAS-based model conceived for modelling the dynamics of systems of cells interacting inside cell environment called niches. (ii) a computational tool, developed for implementing the models and executing the simulations. The tool is meant to work as a kind of virtual laboratory, on top of which kinds of virtual experiments can be performed, characterised by the definition and execution of specific models implemented as MASs, so as to support the validation, falsification and improvement of the models through the observation and analysis of the simulations. A hematopoietic stem cell system is taken as reference case study for formulating a specific model and executing virtual experiments.

Primordial black holes from multi-field inflation

In this master's thesis, the formation of Primordial Black Holes (PBHs) in the context of multi-field inflation is studied. In these models, the interaction of isocurvature and curvature perturbations can lead to a significant enhancement of the latter, and to the subsequent production of PBHs. Depending on their mass, these can account for a significant fraction (or, in some cases, the entirety) of the universe's Dark Matter content. After studying the theoretical framework of generic N-field inflationary models, the focus is restricted to the two-field case, for which a few concrete realisations are analysed. A numerical code (written in Wolfram Mathematica) is developed to make quantitative predictions for the main inflationary observables, notably the scalar power spectra. Parallelly, the production of PBHs due to the dynamics of 2-field inflation is examined: their mass, as well as the fraction of Dark Matter they represent, is calculated for the models considered previously.

Multi-scalar critical theories: first steps for a non perturbative functional renormalization group analysis

In this work the fundamental ideas to study properties of QFTs with the functional Renormalization Group are presented and some examples illustrated. First the Wetterich equation for the effective average action and its flow in the local potential approximation (LPA) for a single scalar field is derived. This case is considered to illustrate some techniques used to solve the RG fixed point equation and study the properties of the critical theories in D dimensions. In particular the shooting methods for the ODE equation for the fixed point potential as well as the approach which studies a polynomial truncation with a finite number of couplings, which is convenient to study the critical exponents. We then study novel cases related to multi field scalar theories, deriving the flow equations for the LPA truncation, both without assuming any global symmetry and also specialising to cases with a given symmetry, using truncations based on polynomials of the symmetry invariants. This is used to study possible non perturbative solutions of critical theories which are extensions of known perturbative results, obtained in the epsilon expansion below the upper critical dimension.

A performance analysis of mesh models for cloud-based workflows

Nell'ambito della loro trasformazione digitale, molte organizzazioni stanno adottando nuove tecnologie per supportare lo sviluppo, l'implementazione e la gestione delle proprie architetture basate su microservizi negli ambienti cloud e tra i fornitori di cloud. In questo scenario, le service ed event mesh stanno emergendo come livelli infrastrutturali dinamici e configurabili che facilitano interazioni complesse e la gestione di applicazioni basate su microservizi e servizi cloud. L’obiettivo di questo lavoro è quello di analizzare soluzioni mesh open-source (istio, Linkerd, Apache EventMesh) dal punto di vista delle prestazioni, quando usate per gestire la comunicazione tra applicazioni a workflow basate su microservizi all’interno dell’ambiente cloud. A questo scopo è stato realizzato un sistema per eseguire il dislocamento di ognuno dei componenti all’interno di un cluster singolo e in un ambiente multi-cluster. La raccolta delle metriche e la loro sintesi è stata realizzata con un sistema personalizzato, compatibile con il formato dei dati di Prometheus. I test ci hanno permesso di valutare le prestazioni di ogni componente insieme alla sua efficacia. In generale, mentre si è potuta accertare la maturità delle implementazioni di service mesh testate, la soluzione di event mesh da noi usata è apparsa come una tecnologia ancora non matura, a causa di numerosi problemi di funzionamento.

Graph-based Vision-Language Model for Multi-Modal Retrieval of Fashion Products

Artificial Intelligence is reshaping the field of fashion industry in different ways. E-commerce retailers exploit their data through AI to enhance their search engines, make outfit suggestions and forecast the success of a specific fashion product. However, it is a challenging endeavour as the data they possess is huge, complex and multi-modal. The most common way to search for fashion products online is by matching keywords with phrases in the product's description which are often cluttered, inadequate and differ across collections and sellers. A customer may also browse an online store's taxonomy, although this is time-consuming and doesn't guarantee relevant items. With the advent of Deep Learning architectures, particularly Vision-Language models, ad-hoc solutions have been proposed to model both the product image and description to solve this problems. However, the suggested solutions do not exploit effectively the semantic or syntactic information of these modalities, and the unique qualities and relations of clothing items. In this work of thesis, a novel approach is proposed to address this issues, which aims to model and process images and text descriptions as graphs in order to exploit the relations inside and between each modality and employs specific techniques to extract syntactic and semantic information. The results obtained show promising performances on different tasks when compared to the present state-of-the-art deep learning architectures.

Hierarchical Multi-Label Text Classification in a Low-Resource Setting

In this thesis we address a multi-label hierarchical text classification problem in a low-resource setting and explore different approaches to identify the best one for our case. The goal is to train a model that classifies English school exercises according to a hierarchical taxonomy with few labeled data. The experiments made in this work employ different machine learning models and text representation techniques: CatBoost with tf-idf features, classifiers based on pre-trained models (mBERT, LASER), and SetFit, a framework for few-shot text classification. SetFit proved to be the most promising approach, achieving better performance when during training only a few labeled examples per class are available. However, this thesis does not consider all the hierarchical taxonomy, but only the first two levels: to address classification with the classes at the third level further experiments should be carried out, exploring methods for zero-shot text classification, data augmentation, and strategies to exploit the hierarchical structure of the taxonomy during training.