Archivio Istituzionale della Ricerca - Università degli Studi di Pavia
Not a member yet
135341 research outputs found
Sort by
Renewable Energy and Self-sufficiency in High-Altitude Alpine Historic Farmstead: A Design Workflow for Optimizing Solar and Water
No full textThe conservation and adaptation of historic buildings in high-altitude environments presents unique challenges, particularly in integrating renewable energy solutions for energy self-sufficiency. The study focuses on the restoration of Monte Fontana Secca and Col de Spadaròt, a 150-hectare alpine farmstead located on the Monte Grappa Massif (Belluno, Italy), and owned by the Fondo per l’Ambiente Italiano ETS (FAI). The site is a typical alpine pasture, without connections to water, electricity, or sewage systems. The intervention aims to revitalize the pasture, reintroduce cattle farming, and restore cheese production. From 2025, the farmstead will open to the public as an educational center dedicated to mountain agriculture and pastoralism, with overnight accommodations. To ensure the site’s energy and water self-sufficiency, the project incorporates local RES and traditional techniques to minimize water and electricity consumption. Active solar systems using copper indium gallium selenide (CIGS) photovoltaic panels are installed on the buildings’ roofs to meet electrical and thermal needs. For water, historical pits (pose) used by herders are restored, and new systems like rainwater harvesting from roofs are implemented. The study outlines the complexity of executing this work at high altitude, detailing the stages of architectural and systems design and construction. The project highlights how combining RES with traditional practices can create low environmental impact and self-sufficient buildings
Adaptive spread of a sexually selected syndrome eliminates an ancient color polymorphism in wall lizards
No full textGenetically determined color morphs are found in many animals. Polymorphism can be maintained by social selection if competitive interactions allow each morph to increase in frequency when rare. This reliance on negative frequency-dependent selection should make color polymorphism vulnerable to the appearance of novel phenotypes that disrupt competitive interactions among morphs. We show that the origin and adaptive spread of a sexually selected syndrome in common wall lizards (Podarcis muralis) selectively eliminates alleles coding for alternative color morphs that have been maintained for millions of years. The results demonstrate how the arrival of a novel phenotype can disrupt balancing selection, providing a link between rapid phenotypic evolution and the loss of color polymorphisms
Applicazioni di Intelligenza Artificiale per la sorveglianza genomica, la modellistica epidemiologica e la predizione evolutiva di SARS-CoV-2
No full textNelle emergenze pandemiche, la sorveglianza genomica è cruciale per l’allerta precoce e l’anticipazione di varianti ad alto potenziale di diffusione; tuttavia la sua efficacia è spesso limitata da tempi di risposta lenti e dall’individuazione tardiva delle varianti, quando queste sono già ampiamente diffuse nella popolazione a rischio. L’intelligenza artificiale (AI) può trasformare la sorveglianza genomica da flusso di reporting retrospettivo a capacità proattiva di supporto alle decisioni. Con l’aumento del sequenziamento e l’evoluzione della diversità virale, l’imprevedibilità deriva da tre fattori principali: eterogeneità e squilibri dei dati, ritardi e lacune nelle conoscenze di riferimento, e assenza di standard condivisi per valutazioni quantitative in contesti operativi reali. Questa tesi affronta tali criticità (i) delimitando gli ambiti in cui i modelli “sequence-aware” sono affidabili, (ii) sviluppando strumenti non supervisionati e generativi per anticipare traiettorie evolutive, e (iii) proponendo rappresentazioni pan-virali riusabili a supporto della sanità pubblica.
Il Capitolo 1 inquadra lo stato dell’arte della sorveglianza genomica basata su AI, evidenziando risultati, problemi aperti e contesto operativo (governance dei dati, privacy, standard di valutazione). Il Capitolo 2 descrive le interazioni tra l’evoluzione della pandemia COVID-19 e la gestione da parte delle autorità sanitarie, mediante un modello autoregressivo su indicatori settimanali europei (inclusi casi COVID-19, ricoveri, decessi, vaccinazioni, varianti, indici di efficienza alla risposta pandemica, come l’indice di stringenza,) e utilizza modelli avanzati di AI, inclusi i foundation models per la previsione a breve termine dell’incidenza. Il Capitolo 3 introduce DeepAutoCoV, un framework non supervisionato per identificare varianti del SARS-CoV-2 che diventeranno Future Dominant Lineages (FDL). Il Capitolo 4 presenta SARITA, un modello autoregressivo domain-adapted per la generazione di sequenze genomiche di SARS-CoV-2: la qualità delle sequenze sintetiche è valutata in base al realismo evolutivo, per esempio la capacità del modello di predire varianti realmente comparse, nonostante non siano state presenti nel dataset di allenamento. Il Capitolo 5 sviluppa MistrVirus, un embedding pan-virale basato sul modello LLM open source dell’azienda Mistral in grado di rappresentare le sequenze genomiche appartenenti a diverse specie di virus in uno spazio multidimensionale; tale rappresentazione può essere usata per allenare modelli di classificazione, per esempio per distinguere i genomi virali dai genomi non virali. Il Capitolo 6 riporta conclusioni complessive, limitazioni e una roadmap verso AI interpretabile e “sequencing-aware” per una sanità pubblica anticipatoria.
Gli studi dei Capitoli 2–5 sono stati condotti presso l’Università di Pavia (Laboratorio di Informatica Biomedica BMI “Mario Stefanelli”) e presso l’Università della Florida, sfruttando infrastrutture HPC (Hipergator) offerte dall’Università della Florida.In pandemic emergencies, genomic surveillance is crucial for early warning and anticipation of variants with high transmission potential; however, its effectiveness is often limited by slow response times and the late detection of variants, when they are already widespread in at-risk populations. Artificial intelligence (AI) can shift genomic surveillance from a retrospective reporting pipeline to a proactive, decision-support capability. As sequencing scales up and viral diversity evolves, unpredictability stems from three main factors: data heterogeneity and imbalance, delays and gaps in reference knowledge, and the lack of shared standards for quantitative evaluation in real-world operational settings. This thesis addresses these challenges by (i) delineating the domains in which “sequence-aware” models are reliable, (ii) developing unsupervised and generative tools to anticipate evolutionary trajectories, and (iii) proposing reusable pan-viral representations to support public-health analytics. Chapter 1 surveys the state of the art in AI-enabled genomic surveillance, highlighting achievements, open problems, and the operational context (data governance, privacy, and evaluation standards). Chapter 2 examines the interplay between the evolution of the COVID-19 pandemic and public-health management through an autoregressive model on European weekly indicators (including COVID-19 cases, hospitalizations, deaths, vaccinations, variants, and indices of pandemic response efficiency, such as the stringency index), and leverages advanced AI—including foundation models—for short-term incidence forecasting. Chapter 3 introduces DeepAutoCoV, an unsupervised framework for identifying SARS-CoV-2 variants that will become Future Dominant Lineages (FDLs). Chapter 4 presents SARITA, a domain-adapted autoregressive model for generating SARS-CoV-2 genomic sequences; the quality of synthetic sequences is assessed via evolutionary realism—for example, the model’s ability to predict variants that later emerged despite being absent from the training set. Chapter 5 develops MistrVirus, a pan-viral embedding based on the Mistral company's open source LLM model capable of representing genomic sequences belonging to different virus species in a multidimensional space; this representation can be used to train classifiers, for instance to distinguish viral from non-viral genomes. Chapter 6 provides overall conclusions, limitations, and a roadmap toward interpretable, “sequence-aware” AI for anticipatory public health. The studies in Chapters 2–5 were conducted at the University of Pavia (Biomedical Informatics Laboratory “Mario Stefanelli”) and the University of Florida, with computation on the HiPerGator HPC infrastructure provided by the University of Florida
Joint deep calibration of the 4-factor PDV model
No full textJoint calibration to SPX and VIX market data is a delicate task that requires sophisticated modeling and incurs high computational costs. The latter is especially true when the pricing of volatility derivatives hinges on nested Monte Carlo simulation. One such example is the 4-factor Markov Path-Dependent Volatility (PDV) model of Guyon and Lekeufack (2023). Nonetheless, its realism has earned it considerable attention in recent years. Gazzani and Guyon (2025) marked a relevant contribution by learning the VIX as a random variable, i.e., a measurable function of the model parameters and the Markovian factors. A neural network replaces the inner simulation, making the joint calibration problem accessible. However, the minimization loop remains slow due to the expensive outer simulation. The present paper overcomes this limitation by learning SPX implied volatilities, VIX futures, and VIX call option prices. The pricing functions reduce to simple matrix–vector products that can be evaluated on the fly, shrinking calibration times to just a few seconds. Notably, we provide standard errors for the optimal calibration parameters
Toxicity of vessel antifouling coating lixiviates in target and non-target marine microalgal species: multi-taxa and biological multi-level approach testing
No full textThe development of a microfilm is one of the very first steps in the succession process, taking place on bare surfaces exposed to aquatic environments, where microalgae, especially diatoms, are among the early colonizers. To prevent or minimize this undesired growth in artificial structures antifouling (AF) measures are applied, coatings being the most common ones. This work studied the effects of two commercially available coatings, a traditional biocide-based one (BC) and an alternative foul-release (FR) one, from a multi-taxa and biological multilevel approach. Three microalgae species were selected, including pelagic non-target species and a benthic target diatom, and exposed for 72 h to AF lixiviates to measure various biological endpoints. Toxicity screening assays revealed that exposure to BC lixiviates inhibited growth in all test species and affected photosynthetic efficiency differently, the diatom being the most sensitive one, while FR lixiviate primarily induced subcellular responses, rather than major physiological impairments. Additionally, exposure altered total pigment content in the three tested algae, particularly under BC treatments. Subcellular responses demonstrated differences in biomarkers of oxidative stress (catalase and glutathione S-transferase activities and lipid peroxidation levels) in the microalgae between BC and FR treatments. Cellular metal sorption levels did not show clear differences across treatments nor species. Overall, exposure to BC and FR antifouling lixiviates directly affected both target and non-target microalgae species, although the type and magnitude of the responses varied according to species and treatments. Multi-taxa and multi-level approaches with microalgae provide a broad overview of the biological responses and serve as a valuable tool in aquatic toxicology
From the Shadow of Universal History. Finis Terrae and Liberation in the Thought of Amelia Podetti and Enrique Dussel
No full textThis paper explores the respective interpretations of Hegel’s philosophy of history by Argentine thinkers Amelia Podetti and Enrique Dussel. Although they likely met, they shared a similar intellectual project, characterized by a critical appropriation of European thought, aiming to incorporate Latin America’s role in world history. For Dussel 1492 serves as the original and constitutive milestone of Modernity, and he argues that the Cartesian ego cogito is founded upon – and inseparable from – an ego conquiro that implies the hiding and expulsion of the Other. Podetti, conversely, asserts that only from Latin America, as finis terrae, is it “possible to perceive in its true form and dimensions the history of man on the planet”. Both Podetti and Dussel offer disobedient readings of Hegel’s philosophy. They strategically appropriate ideas and conceptual matrices from the philosopher of the Phenomenology with the aim to subvert his philosophical project and rewrite it, synthesizing it with entirely different paradigms of thought and novel philosophical-political experiences
«La composta bellezza del ragionamento strenuo e serrato». Sulle forme del saggio in Primo Levi
No full textIl saggio prende in esame il saggismo di Primo Levi, individuandone le diverse forme e analizzandone in particolare le scelte linguistiche e argomentative. Il lavoro procede per focalizzazioni progressive e si chiude con un esercizio di analisi stilistica su di un campione privilegiato: "François Rabelais", un elzeviro risalente al 1964 e da Levi poi incluso con questo titolo nella raccolta saggistica dell’"Altrui mestiere" (1985)
Prevalence of vaccine hesitancy in Italy: a cross-sectional study
No full textBackground
Vaccine hesitancy (VH) remains a global threat, exacerbated by socio-political uncertainty. We aimed primarily to estimate VH prevalence in Italy, identifying the most susceptible subgroups, and secondarily to assess whether these patterns varied across VH dimensions.
Methods
Cross-sectional survey (web/telephone) among adults in Italy (September 2024–March 2025). The sample (n = 52,094) was nationally representative by age, gender, education, area, municipality size. The primary outcome was VH (score ≥25, adult Vaccine Hesitancy Scale, aVHS). The secondary outcomes were aVHS subscales “Lack of trust” and “Risk perception”. Post-stratification weighting for age, area, and municipality size was applied.
Findings
VH prevalence was 46.09% (95% CI: 45.65–46.53%). Multivariable models showed several associations with VH, e.g., gender, sexual orientation, ethnicity, health literacy, political and religious orientation, personal experiences, and vaccination support from community figures. Among many subgroups significant after multiple-comparison correction, the strongest differences in VH predicted probability (PP) were estimated among individuals using complementary/alternative medicine (PP = 58.5%), right-aligned (PP = 47.0%) or politically unaffiliated participants (PP = 48.4%), individuals with middle school education (PP = 48.3%), people aged 60–74 (PP = 49.0%), and participants uncertain about healthcare workers' pro-vaccination support (PP = 52.8%). While some groups, e.g., individuals with chronic conditions, inadequate health literacy, or religious participants reported higher perceived risk, others, e.g., non-binary respondents, showed higher lack of trust.
Interpretation
This study highlighted the importance of granular data to inform inclusive strategies. Key figures and politics emerged as relevant, deserving further exploration. Future research should evaluate tailored interventions for identified at-risk groups.
Funding
NextGenerationEU funding within the Italian Ministry of University and Research PNRR Extended Partnership initiative on Emerging Infectious Diseases