Un sistema informatico per la gestione di patologie oncologiche croniche

L'assistenza sanitaria in Italia e nel mondo è caratterizzata da bisogni elevati in continua crescita, ad essi si contrappone l'attuale crisi delle risorse economiche determinando per il Sistema una valutazione di scelte quali la riduzione o la rimodulazione dell'offerta sanitaria pubblica. L'idea di questo lavoro, nata all'interno dell'Istituto Scientifico Romagnolo per lo Studio e la Cura dei Tumori (IRST) IRCCS, è di approcciare questo problema in ottica di miglioramento delle performance anziché riduzione dei servizi, nella convinzione che vi siano importanti margini di perfezionamento. Per questi motivi si è valutata la necessità di sviluppare un metodo e un'applicazione software per l'identificazione dei percorsi diagnostici terapeutici assistenziali (PDTA), per la raccolta di dati dalle strutture coinvolte, per l'analisi dei costi e dei risultati, mirando ad una analisi di tipo costi - efficacia e di benchmarking in ottica di presa in carico dei bisogni di salute. La tesi descrive la fase di raccolta e analisi dei requisiti comprensiva della profilazione utenti e descrizione di alcuni aspetti dinamici salienti, la fase di progettazione concettuale (schema Entity/Relationship, glossario e volumi dei dati), la fase di progettazione logica e la prototipazione dell'interfaccia utente. Riporta inoltre una valutazione dei tempi di sviluppo realizzata tramite metodologia di calcolo dei punti per caso d'uso. L'applicazione progettata è oggetto di valutazione di fattibilità presso l'IRST, che ha utilizzato alcune delle metodologie descritte nella tesi per descrivere il percorso di patologia mammaria e presentarne i primi risultati all'interno di un progetto di ricerca in collaborazione con l'Agenzia Nazionale per i Servizi Sanitari Regionali (Agenas).

Organic matter deposition/resuspension in a one-dimensional physical-biogeochemical model. A modelling study

The shallow water configuration of the gulf of Trieste allows the propagation of the stress due to wind and waves along the whole water column down to the bottom. When the stress overcomes a particular threshold it produces resuspension processes of the benthic detritus. The benthic sediments in the North Adriatic are rich of organic matter, transported here by many rivers. This biological active particulate, when remaining in the water, can be transported in all the Adriatic basin by the basin-wide circulation. In this work is presented a first implementation of a resuspension/deposition submodel in the oceanographic coupled physical-biogeochemical 1-dimensional numerical model POM-BFM. At first has been considered the only climatological wind stress forcing, next has been introduced, on the surface, an annual cycle of wave motion and finally have been imposed some exceptional wave event in different periods of the year. The results show a strong relationship between the efficiency of the resuspension process and the stratification of the water column. During summer the strong stratification can contained a great quantity of suspended matter near to the bottom, while during winter even a low concentration of particulate can reach the surface and remains into the water for several months without settling and influencing the biogeochemical system. Looking at the biologic effects, the organic particulate, injected in the water column, allow a sudden growth of the pelagic bacteria which competes with the phytoplankton for nutrients strongly inhibiting its growth. This happen especially during summer when the suspended benthic detritus concentration is greater.

Georges Lemaître: a visionary contribute to the quest of the world model

The aim of this dissertation is to present the sequence of events which brought the scientific community of the early 20th century to conceive an expanding Universe born from a single origin. Among the facts here reported, some are well-known, some others instead are little-known backstories, not so easy neither to obtain nor to trust. Indeed, several matters shown in this thesis, now as then, create a battleground among scientists. Amid the numerous personalities whose contributions are discussed in this work, the main protagonist is surely Georges Lemaître, who managed to combine – without overlapping – his being both a priest and a scientist. The first chapter is dedicated to his biography, from his childhood in Belgium, to his early adulthood between England and the USA, to his success in the scientific community. The second and the third chapter explain how the race to the understanding of a Universe which not only expands, but also originated from a singularity, developed. The Belgian priest’s discoveries, as shown, were challenged by other important scientists, who, in several cases, Lemaître had a friendly relationship with. As a consequence, the fourth and final chapter deals with the multiple relations that the priest managed to build, thanks to his politeness and kindness. Moreover, it is also covered Lemaître’s personal connection with the Church and religion, without forgetting the personalities that influenced him – above all, Saint Thomas Aquinas. As a conclusion to this thesis, two appendices gather not only a summary of Lemaître’s works which are not already described in the chapters, but also the biographies of all the characters presented in this dissertation.

Vocabulary Transfer and Knowledge Distillation for Language Model Compression

Natural Language Processing (NLP) has seen tremendous improvements over the last few years. Transformer architectures achieved impressive results in almost any NLP task, such as Text Classification, Machine Translation, and Language Generation. As time went by, transformers continued to improve thanks to larger corpora and bigger networks, reaching hundreds of billions of parameters. Training and deploying such large models has become prohibitively expensive, such that only big high tech companies can afford to train those models. Therefore, a lot of research has been dedicated to reducing a model’s size. In this thesis, we investigate the effects of Vocabulary Transfer and Knowledge Distillation for compressing large Language Models. The goal is to combine these two methodologies to further compress models without significant loss of performance. In particular, we designed different combination strategies and conducted a series of experiments on different vertical domains (medical, legal, news) and downstream tasks (Text Classification and Named Entity Recognition). Four different methods involving Vocabulary Transfer (VIPI) with and without a Masked Language Modelling (MLM) step and with and without Knowledge Distillation are compared against a baseline that assigns random vectors to new elements of the vocabulary. Results indicate that VIPI effectively transfers information of the original vocabulary and that MLM is beneficial. It is also noted that both vocabulary transfer and knowledge distillation are orthogonal to one another and may be applied jointly. The application of knowledge distillation first before subsequently applying vocabulary transfer is recommended. Finally, model performance due to vocabulary transfer does not always show a consistent trend as the vocabulary size is reduced. Hence, the choice of vocabulary size should be empirically selected by evaluation on the downstream task similar to hyperparameter tuning.

Meta-learning for informed model searches in Automated Machine Learning

Day by day, machine learning is changing our lives in ways we could not have imagined just 5 years ago. ML expertise is more and more requested and needed, though just a limited number of ML engineers are available on the job market, and their knowledge is always limited by an inherent characteristic of theirs: they are humans. This thesis explores the possibilities offered by meta-learning, a new field in ML that takes learning a level higher: models are trained on other models' training data, starting from features of the dataset they were trained on, inference times, obtained performances, to try to understand the relationship between a good model and the way it was obtained. The so-called metamodel was trained on data collected by OpenML, the largest ML metadata platform that's publicly available today. Datasets were analyzed to obtain meta-features that describe them, which were then tied to model performances in a regression task. The obtained metamodel predicts the expected performances of a given model type (e.g., a random forest) on a given ML task (e.g., classification on the UCI census dataset). This research was then integrated into a custom-made AutoML framework, to show how meta-learning is not an end in itself, but it can be used to further progress our ML research. Encoding ML engineering expertise in a model allows better, faster, and more impactful ML applications across the whole world, while reducing the cost that is inevitably tied to human engineers.

Development of a relationship between physical parameters of tsunami waves and tsunami intensity based on macroscopic effects and its application to the Italian tsunami database

Tsunamis are rare events. However, their impact can be devastating and it may extend to large geographical areas. For low-probability high-impact events like tsunamis, it is crucial to implement all possible actions to mitigate the risk. The tsunami hazard assessment is the result of a scientific process that integrates traditional geological methods, numerical modelling and the analysis of tsunami sources and historical records. For this reason, analysing past events and understanding how they interacted with the land is the only way to inform tsunami source and propagation models, and quantitatively test forecast models like hazard analyses. The primary objective of this thesis is to establish an explicit relationship between the macroscopic intensity, derived from historical descriptions, and the quantitative physical parameters measuring tsunami waves. This is done first by defining an approximate estimation method based on a simplified 1D physical onshore propagation model to convert the available observations into one reference physical metric. Wave height at the coast was chosen as the reference due to its stability and independence of inland effects. This method was then implemented for a set of well-known past events to build a homogeneous dataset with both macroseismic intensity and wave height. By performing an orthogonal regression, a direct and invertible empirical relationship could be established between the two parameters, accounting for their relevant uncertainties. The target relationship is extensively tested and finally applied to the Italian Tsunami Effect Database (ITED), providing a homogeneous estimation of the wave height for all existing tsunami observations in Italy. This provides the opportunity for meaningful comparison for models and simulations, as well as quantitatively testing tsunami hazard models for the Italian coasts and informing tsunami risk management initiatives.

Assessing the relationship between bee/flower diversity and vegetation structural heterogeneity through UAV photogrammetry

The ecosystem services provided by bees are very important. Factors as habitat fragmentation, intensive agriculture and climate change are contributing to the decline of bee populations. The use of remote sensing could be a useful tool for the recognition of sites with a high diversity, before performing a more expensive survey in the field. In this study the ability of Unmanned Aerial Vehicles (UAV) images to estimate biodiversity at local scale has been analysed testing the concept of the Height Variation Hypothesis (HVH). This approach states that, the higher the vegetation height heterogeneity (HH) measured by remote sensing information, the higher the vertical complexity and the higher vegetation species diversity. In this thesis the concept has been brought to a higher level, in order to understand if the vegetation HH can be considered a proxy also of bee species diversity and abundance. We tested this approach collecting field data on bees/flowers and RGB images through an UAV campaign in 30 grasslands in the South of the Netherlands. The Canopy Height Model (CHM) were derived through the photogrammetry technique "Structure from Motion" (SfM) with resolutions of 10cm, 25cm, 50cm. Successively, the HH assessed on the CHM using the Rao's Q heterogeneity index was correlated to the field data (bee abundance, diversity and bee/flower species richness). The correlations were all positive and significant. The highest R2 values were found when the HH was calculated at 10cm and correlated to bee species richness (R2 = 0.41) and Shannon’s H index (R2 = 0.38). Using a lower spatial resolution the goodness of fit slightly decreases. For flower species richness the R2 ranged between 0.36 to 0.39. Our results suggest that methods based on the concept behind the HVH, in this case deriving information of HH from UAV data, can be developed into valuable tools for large-scale, standardized and cost-effective monitoring of flower diversity and of the habitat quality for bees.