Landslide susceptibility in the Belt and Road Initiative

The Belt and Road Initiative (BRI) is a project launched by the Chinese Government whose main goal is to connect more than 65 countries in Asia, Europe, Africa and Oceania developing infrastructures and facilities. To support the prevention or mitigation of landslide hazards, which may affect the mainland infrastructures of BRI, a landslide susceptibility analysis in the countries involved has been carried out. Due to the large study area, the analysis has been carried out using a multi-scale approach which consists of mapping susceptibility firstly at continental scale, and then at national scale. The study area selected for the continental assessment is the south-Asia, where a pixel-based landslide susceptibility map has been carried out using the Weight of Evidence method and validated by Receiving Operating Characteristic (ROC) curves. Then, we selected the regions of west Tajikistan and north-east India to be investigated at national scale. Data scarcity is a common condition for many countries involved into the Initiative. Therefore in addition to the landslide susceptibility assessment of west Tajikistan, which has been conducted using a Generalized Additive Model and validated by ROC curves, we have examined, in the same study area, the effect of incomplete landslide dataset on the prediction capacity of statistical models. The entire PhD research activity has been conducted using only open data and open-source software. In this context, to support the analysis of the last years an open-source plugin for QGIS has been implemented. The SZ-tool allows the user to make susceptibility assessments from the data preprocessing, susceptibility mapping, to the final classification. All the output data of the analysis conducted are freely available and downloadable. This text describes the research activity of the last three years. Each chapter reports the text of the articles published in international scientific journal during the PhD.

Numerical and experimental analysis of 3D printed smart structures and systems

Three dimensional (3D) printers of continuous fiber reinforced composites, such as MarkTwo (MT) by Markforged, can be used to manufacture such structures. To date, research works devoted to the study and application of flexible elements and CMs realized with MT printer are only a few and very recent. A good numerical and/or analytical tool for the mechanical behavior analysis of the new composites is still missing. In addition, there is still a gap in obtaining the material properties used (e.g. elastic modulus) as it is usually unknown and sensitive to printing parameters used (e.g. infill density), making the numerical simulation inaccurate. Consequently, the aim of this thesis is to present several work developed. The first is a preliminary investigation on the tensile and flexural response of Straight Beam Flexures (SBF) realized with MT printer and featuring different interlayer fiber volume-fraction and orientation, as well as different laminate position within the sample. The second is to develop a numerical analysis within the Carrera' s Unified Formulation (CUF) framework, based on component-wise (CW) approach, including a novel preprocessing tool that has been developed to account all regions printed in an easy and time efficient way. Among its benefits, the CUF-CW approach enables building an accurate database for collecting first natural frequencies modes results, then predicting Young' s modulus based on an inverse problem formulation. To validate the tool, the numerical results are compared to the experimental natural frequencies evaluated using a digital image correlation method. Further, we take the CUF-CW model and use static condensation to analyze smart structures which can be decomposed into a large number of similar components. Third, the potentiality of MT in combination with topology optimization and compliant joints design (CJD) is investigated for the realization of automated machinery mechanisms subjected to inertial loads.

Machine learning "as a service" for high energy physics (MLaaS4HEP): evolution of a framework for ML-based physics analyses

The scientific success of the LHC experiments at CERN highly depends on the availability of computing resources which efficiently store, process, and analyse the amount of data collected every year. This is ensured by the Worldwide LHC Computing Grid infrastructure that connect computing centres distributed all over the world with high performance network. LHC has an ambitious experimental program for the coming years, which includes large investments and improvements both for the hardware of the detectors and for the software and computing systems, in order to deal with the huge increase in the event rate expected from the High Luminosity LHC (HL-LHC) phase and consequently with the huge amount of data that will be produced. Since few years the role of Artificial Intelligence has become relevant in the High Energy Physics (HEP) world. Machine Learning (ML) and Deep Learning algorithms have been successfully used in many areas of HEP, like online and offline reconstruction programs, detector simulation, object reconstruction, identification, Monte Carlo generation, and surely they will be crucial in the HL-LHC phase. This thesis aims at contributing to a CMS R&D project, regarding a ML "as a Service" solution for HEP needs (MLaaS4HEP). It consists in a data-service able to perform an entire ML pipeline (in terms of reading data, processing data, training ML models, serving predictions) in a completely model-agnostic fashion, directly using ROOT files of arbitrary size from local or distributed data sources. This framework has been updated adding new features in the data preprocessing phase, allowing more flexibility to the user. Since the MLaaS4HEP framework is experiment agnostic, the ATLAS Higgs Boson ML challenge has been chosen as physics use case, with the aim to test MLaaS4HEP and the contribution done with this work.

Modelli di Machine Learning per la Valutazione della Qualità del Vino

Descrizione, implementazione in Python e valutazione di modelli di Machine Learning e di tutte le sue fasi di Preprocessing, EDA, Training, Test e Evaluation, per valutare la qualità del vino attraverso le sue caratteristiche fisico-chimiche.

ICE Emissions Estimation through Machine-learning techniques to support real-world R&D experiments

The emissions estimation, both during homologation and standard driving, is one of the new challenges that automotive industries have to face. The new European and American regulation will allow a lower and lower quantity of Carbon Monoxide emission and will require that all the vehicles have to be able to monitor their own pollutants production. Since numerical models are too computationally expensive and approximated, new solutions based on Machine Learning are replacing standard techniques. In this project we considered a real V12 Internal Combustion Engine to propose a novel approach pushing Random Forests to generate meaningful prediction also in extreme cases (extrapolation, very high frequency peaks, noisy instrumentation etc.). The present work proposes also a data preprocessing pipeline for strongly unbalanced datasets and a reinterpretation of the regression problem as a classification problem in a logarithmic quantized domain. Results have been evaluated for two different models representing a pure interpolation scenario (more standard) and an extrapolation scenario, to test the out of bounds robustness of the model. The employed metrics take into account different aspects which can affect the homologation procedure, so the final analysis will focus on combining all the specific performances together to obtain the overall conclusions.

Bio-QA-GNN: Reasoning with Language Models and Knowledge Graphs for Interpretable Biomedical Question Answering

Nowadays the idea of injecting world or domain-specific structured knowledge into pre-trained language models (PLMs) is becoming an increasingly popular approach for solving problems such as biases, hallucinations, huge architectural sizes, and explainability lack—critical for real-world natural language processing applications in sensitive fields like bioinformatics. One recent work that has garnered much attention in Neuro-symbolic AI is QA-GNN, an end-to-end model for multiple-choice open-domain question answering (MCOQA) tasks via interpretable text-graph reasoning. Unlike previous publications, QA-GNN mutually informs PLMs and graph neural networks (GNNs) on top of relevant facts retrieved from knowledge graphs (KGs). However, taking a more holistic view, existing PLM+KG contributions mainly consider commonsense benchmarks and ignore or shallowly analyze performances on biomedical datasets. This thesis start from a propose of a deep investigation of QA-GNN for biomedicine, comparing existing or brand-new PLMs, KGs, edge-aware GNNs, preprocessing techniques, and initialization strategies. By combining the insights emerged in DISI's research, we introduce Bio-QA-GNN that include a KG. Working with this part has led to an improvement in state-of-the-art of MCOQA model on biomedical/clinical text, largely outperforming the original one (+3.63\% accuracy on MedQA). Our findings also contribute to a better understanding of the explanation degree allowed by joint text-graph reasoning architectures and their effectiveness on different medical subjects and reasoning types. Codes, models, datasets, and demos to reproduce the results are freely available at: \url{https://github.com/disi-unibo-nlp/bio-qagnn}.

Il segnale EEG in un compito di ’center-out reaching’: analisi e decodifica a livello di scalpo e di corteccia

Nellâ analisi del segnale EEG, sono di particolare interesse i potenziali evento correlato (ERP), che descrivono la risposta cerebrale in seguito alla presentazione di stimoli o eventi esterni o interni. Questi potenziali non sono immediatamente visibili sul tracciato EEG grezzo, ed è necessario, oltre ad unâ��accurata fase di preprocessing, mediare (averaging) i segnali di molti trial ripetuti per visualizzare tali risposte nel tempo. Questo studio ha posto l' attenzione sugli ERP visuomotori generati in un compito di center-out reaching, che prevede il raggiungimento di uno tra cinque target, ognuno associato ad un LED, mediante il braccio dominante, con una tempistica scandita dalla presentazione di due stimoli visivi: lo stimolo preparatorio ¸ (che indica il target) e lo stimolo imperativo (che dà il via libera al movimento). I segnali ERP, ottenuti mediante la tecnica dellâ averaging, sono stati analizzati sia a livello di scalpo, considerando i segnali di elettrodo, sia a livello di corteccia, dopo risoluzione del problema inverso, e considerando rappresentazioni prima a livello di singoli dipoli corticali e quindi di intere regioni corticali (ROI). Inoltre, è stata applicato un metodo di deep learning (rete neurale convoluzionale) per decodificare il segnale EEG a livello di singolo trial, ovvero classificare il target coinvolto nello specifico trial. La decodifica è stata applicata sia ai segnali di scalpo sia ai segnali delle ROI corticali. Complessivamente i risultati mostrano ERP ben visibili a livello di scalpo e legati sia a processing visivo che motorio. Gli ERP a livello di ROI corticali sono più rumorosi e sembrano cogliere meno processing motorio rispetto al visivo, presumibilmente anche in conseguenza di alcune scelte metodologiche nella ricostruzione di segnali di ROI. In linea con questo, le performance di decodifica sono migliori a livello di scalpo che di ROI corticali.