186698 research outputs found
Sort by
New Approaches of eXplainable AI: From Video Analytics to Federated Learning
No full textQuesta tesi indaga come l’eXplainable Artificial Intelligence (XAI) possa essere sistematicamente incorporata nelle pipeline di Machine Learning (ML) per applicazioni in scenari safety–critical. L’argomento centrale è che l’explainability non debba essere trattata come una funzionalità post-hoc, ma come un principio di progettazione che guida la costruzione, il monitoraggio e il dispiegamento dei sistemi. Il lavoro si sviluppa in due domini: la video analytics per la mobilità autonoma e assistiva, e l’apprendimento distribuito in presenza di vincoli di privacy e proprietà dei dati.
Nel contesto della video analytics, sono state definite solide basi addestrando e ottimizzando YOLOv8s per ambienti indoor, con elevate prestazioni nel rilevamento di
persone e sedie a rotelle. Su questo backbone è stato sviluppato un Operational Design Domain (ODD) Checker, che combina l’analisi delle feature visive con regole basate su
alberi di decisione (DT) per un monitoraggio interpretabile e verificabile della sicurezza. L’explainability è stata poi estesa al ragionamento a livello di scena tramite una valutazione comparativa di modelli Vision–Language (CLIP, MiniGPT-4, GPT-4V), capaci di classificare ambienti di navigazione come “Safe to Proceed” o “Risky to Proceed” e
contestualizzati nel paradigma dei cigni per la gestione dei rischi rari. Per quantificare l’affidabilità predittiva, la Conformal Prediction (CP) è stata applicata al rilevamento di oggetti, fornendo garanzie statistiche finite e mettendo in luce i compromessi tra strategie di
quantificazione dell’incertezza (Uncertainty Quantification, UQ) box-wise e image-wise.
Nel dominio dell’apprendimento distribuito, la tesi introduce Federated Learning with Interpretable Rule Transfer (FL-IRT), un framework che sostituisce l’aggregazione opaca
dei parametri con la costruzione di modelli basati su regole sia lato client che lato server. FL-IRT consente di ottenere modelli globali competitivi in termini di accuratezza ma anche trasparenti nel processo decisionale, supportando al contempo meccanismi di aggregazione sicura e conformità con la normativa GDPR. Esperimenti condotti su diversi dataset confermano la sua scalabilità, la robustezza in condizioni non-iid e un notevole miglioramento di efficienza rispetto ai baseline neurali.
Complessivamente, questi contributi dimostrano che l’explainability può essere integrata a diversi livelli di astrazione—dalle feature pixel–level e dal rilevamento di oggetti, fino al ragionamento semantico, alla calibrazione statistica e all’apprendimento distribuito. Avanzando la video analytics interpretabile, la quantificazione dell’incertezza basata su principi formali e i framework federati trasparenti, la tesi mostra che l’AI trustworthy-by-design è realizzabile senza sacrifici proibitivi in termini di accuratezza o efficienza. L’implicazione più ampia è che la XAI funge da livello regolatorio nell’AI, trasformando principi astratti di responsabilità e sicurezza in standard ingegneristici applicabili.This thesis investigates how eXplainable Artificial Intelligence (XAI) can be systematically embedded into Machine Learning (ML) pipelines for safety–critical applications. The central argument is that explainability should not be treated as a post-hoc feature but as a design principle that governs how systems are constructed, monitored, and deployed. The work spans two domains: video analytics for autonomous and assistive mobility, and distributed learning under privacy and ownership constraints.
On the video analytics side, strong baselines were established by fine-tuning YOLOv8s for indoor mobility, achieving high accuracy in detecting people and wheelchairs. Building on this backbone, an Operational Design Domain (ODD) Checker was introduced, combining visual feature analysis with Decision Tree (DT) rules to provide interpretable, auditable safety monitoring. Explainability was extended to scene-level reasoning through benchmarking of Vision–Language Models (CLIP, MiniGPT-4, GPT-4V), which were evaluated on their ability to classify navigation scenes as “Safe to Proceed” or “Risky to Proceed” and contextualized within the swan metaphor for rare risks. To quantify predictive reliability, Conformal Prediction (CP) was applied to object detection, establishing finite-sample coverage guarantees and demonstrating the trade-offs between box-wise and image-wise Uncertainty Quantification (UQ) strategies.
In distributed learning, the thesis introduces Federated Learning with Interpretable Rule Transfer (FL-IRT), a framework that replaces opaque parameter averaging with the construction of rule-based models at client and server levels. FL-IRT enables global models that are both competitive in accuracy and transparent in logic, while supporting secure aggregation and GDPR-compliant privacy mechanisms. Experimental results across multiple datasets confirm its scalability, robustness to non-iid conditions, and significant efficiency gains over neural baselines.
Taken together, these contributions show that explainability can be embedded across abstraction layers—from pixel-level features and object detection to semantic reasoning, statistical calibration, and distributed learning. By advancing interpretable video analytics, principled UQ, and transparent federated frameworks, the thesis demonstrates
that trustworthy-by-design AI is achievable without prohibitive sacrifices in accuracy or efficiency. The broader implication is that XAI functions as a regulatory layer in AI, transforming abstract principles of accountability and safety into enforceable engineering standards
Self-avoiding closed curves in the regular and semiregular grids
No full textWe consider closed curves in the three regular and eight semiregular grids in the plane, in which each vertex and each edge can be repeated a limited number of times. We define the conditions for such curves to be self-avoiding, and we present a linear-time algorithm to check them. We define the orientation of such curves. We propose a classification of their vertices, and we give a unifying formula relating the number of different types of vertices, valid in the regular and semiregular grids. Our results can be used in the plane tiling applications
"Fiore dei conversi". Testo inedito di fine secolo XIV
No full text"Fiore dei conversi" is an anonymous poem in refrained sonnets (372 poems) dating back to the end of the fourteenth century and preserved in a single manuscript (Vatican Apostolic Library, Chig. M.IV.99). It recounts an allegorical journey among the stars and to Paradise undertaken by the protagonist while still alive, accompanied by a wise and faithful guide, often referred to as the 'duca'. This work presents the critical edition of the poem: the first part (chapters 1‐6) offers a general introduction to the text from
a literary and linguistic perspective, and includes a codicological description of the manuscript. Within this section, one chapter is devoted to the influence of Dante’s "Commedia" on the imagination, linguistic 'usus scribendi', and stylistic register of the anonymous poet. Furthermore, the study advances the hypothesis of a geographical localisation of the poem on the basis of several relevant linguistic features, whose presence suggests an origin in central Italy, particularly in Umbria. The second section of the work (chapter 7) contains the critical edition of the sonnets: each poem is
preceded by an introduction summarising its content and followed by a commentary highlighting its most notable literary aspects. The poem may be considered one of the most interesting and innovative examples of the reception and reworking of Dante’s masterpiece, the "Commedia"
Evaluating the robustness of explainable AI in medical image recognition under natural and adversarial data corruption
No full textThe integration of Explainable AI (XAI) into healthcare promises greater transparency and interpretability of machine learning models, enabling clinicians to understand predictions and make more reliable medical decisions. Yet, the robustness of XAI methods remains uncertain, as small input perturbations can drastically change their explanations, posing critical risks in clinical settings where they may lead to misdiagnoses or inappropriate treatment. Motivated by the central role of XAI in healthcare decision-making, this paper examines its robustness in the presence of data corruption. We systematically evaluate the stability of widely used XAI techniques against both naturally occurring noise (e.g., JPEG compression) and adversarial manipulations that alter explanations without affecting model predictions. To this end, we introduce a set of evaluation metrics that capture complementary aspects of explanation stability, ranging from pixel-level consistency to spatial coherence, and propose a protocol for assessing the resilience of XAI methods across diverse perturbation sources. Our analysis spans three medical imaging datasets, various convolutional and transformer models, and ten post-hoc XAI methods, including Grad-CAM++ for convolutional networks and LibraGrad for vision transformers. We find that current XAI techniques are often unstable, even under imperceptible perturbations. For adversarial noise, a clear set of robust methods emerges, whereas for natural noise, performance varies, with some methods maintaining spatial stability and others preserving pixel-wise consistency. All results together highlight the need for multi-perspective evaluation when selecting XAI techniques in practice
Experimental investigation of bow-mounted passive and active hydrofoils for ship resistance reduction in regular waves
No full textThis study experimentally investigates the influence of passive and active bow-mounted hydrofoils on ship-resistance reduction and thrust generation in regular waves. Towing-tank experiments were performed on a Series 60 ship model equipped with both passive and active hydrofoils to quantify their effects on the total resistance. The active hydrofoil system, actuated by a servomechanism, was continuously adjusted in angle of attack in synchronization with the bow's vertical motion, whereas the passive hydrofoil was kept at a fixed angle of attack of zero degrees. Tests were conducted over various forward speeds, wavelength-to-ship-length ratios, and active-hydrofoil oscillation amplitudes to evaluate the influence of these parameters on hydrodynamic performance. The results indicate that the active hydrofoil generated a notable propulsive thrust, achieving up to a 32 % reduction in the mean total resistance compared with the passive configuration. These findings emphasize the superior energy-harvesting potential of active hydrofoils and their promise for reducing ship power demand and fuel consumption. The results support the design and control optimization of wave-interaction devices for ships
Imprecise model of thunderstorm wind speed and uncertainty propagation on the maximum dynamic response
No full textThe wind velocity field associated with thunderstorm downbursts can be modelled as a uniformly modulated nonstationary random process, characterized by an Evolutionary Power Spectral Density function. The parameters characterizing the evolutionary model vary significantly from one thunderstorm to another. Due to the limited data availability, the interval model appears to be a suitable approach to represent the uncertainty of such parameters. In this paper, by leveraging available thunderstorm data and a literature-based model for the vertical profile of mean velocity, appropriate bounds for the key loading parameters are established, and an interval model for the thunderstorm wind speed is introduced. Employing a closed-form solution for the gust response factor and based on the introduced interval model of the thunderstorm wind speed, this study investigates the propagation of uncertainties on the thunderstorm gust response factor and the maximum dynamic response for slender vertical structures using the Improved Interval Analysis. Results indicate that, for the structural cases analyzed, uncertainties in thunderstorm parameters exert a more significant influence on the thunderstorm gust response factor and maximum response than those in the structural parameters
“WHY ARE WE FORCED TO TALK ABOUT INCLUSION AND INTERCULTURALITY?” A Critical Research Journey toward Justice and Equity in Early Childhood Education in Italy: A Mixed-Methods Collection of Studies.
No full textThis doctoral dissertation critically examines interculturality and inclusion in public early childhood education (ECE) services (ages 0–6) in the Northwest Italy. While initially framed within mainstream institutional approaches to intercultural competences and educator well-being, the research progressively revealed the limits—and risks—of treating interculturality as a technical, individual, or depoliticized construct. Rather than confirming pre-existing models, the research process itself generated a profound theoretical, methodological, and ethical shift: interculturality emerged as intelligible only when situated within its structural, political, and epistemic conditions.
Developed through an institutional collaboration between the University of Genoa and the Municipality of Genoa and funded by Italy’s National Recovery and Resilience Plan (PNRR), the thesis includes a mixed-methods design articulated across four interconnected empirical studies. An initial quantitative phase explores the relationship between educators’ intercultural competences, well-being, and burnout. Subsequent qualitative, ethnographic, and narrative studies foreground the everyday experiences of educators and parents with migration backgrounds, tracing how interculturality and inclusion are negotiated within asymmetrical institutional, labor, and policy contexts.
A central contribution of the dissertation lies in its reflexive and critical epistemological stance. Drawing on liberation psychology, feminist and intersectional theories, decolonial thought, and narrative practice, the research interrogates how dominant constructs—such as burnout, intercultural competences, and inclusion—can individualize systemic injustice and obscure relations of power. Auto-reflexivity is treated not as a methodological add-on but as an ethical and political practice that exposes the researcher’s positionality, institutional constraints, and complicity within knowledge production.
Findings demonstrate that interculturality cannot be reduced to skill acquisition, diversity management, or symbolic inclusion. Instead, it is a relational and contested process shaped by structural precarity, labor conditions, governance arrangements, and broader regimes of bordering and exclusion. Educators’ well-being emerges as a matter of social justice rather than individual resilience, while parents with history of migration appear as active epistemic agents and co-constructors of educational spaces.
The dissertation ultimately argues that speaking of interculturality without addressing its systemic and political foundations risks reproducing the very inequalities it claims to challenge. By documenting the passage from critical awareness to concrete transformations in school routines, spaces, and decision-making processes, the study shows how early childhood education can function as a strategic site for epistemic justice and structural change. These transformations demonstrate that inclusion becomes meaningful only when it is enacted through participatory, decolonial, and institutionally accountable practices embedded in everyday educational life
Antibiotic Resistance in Ligurian Fauna
No full textAntimicrobial resistance (AMR) represents one of the most critical global public health challenges of the twenty-first century, posing a serious threat to human, animal, and environmental health. The dissemination of resistant bacteria and resistance genes is increasingly recognized as a complex phenomenon driven by interactions among multiple ecological compartments, highlighting the need for integrated surveillance approaches within the One Health framework.
This doctoral thesis investigates the role of wildlife and the environment as reservoirs and indicators of antimicrobial resistance in the Ligurian region (North-West Italy), an area characterized by high anthropogenic pressure and close interactions between urban, agricultural, terrestrial, and marine ecosystems.
The first part of the thesis provides an extensive overview of the biological mechanisms underlying antimicrobial resistance, including enzymatic inactivation, target modification, efflux systems, and horizontal gene transfer. Particular attention is given to the environmental dimension of AMR and to the growing evidence supporting the role of natural ecosystems in the persistence and dissemination of resistant bacteria.
The experimental section includes two complementary investigations. The first study focuses on the detection and characterization of Shiga toxin-producing Escherichia coli (STEC) isolated from wild ungulates in Liguria. Molecular analyses revealed the presence of virulence-associated genes, confirming the circulation of potentially pathogenic strains in wildlife populations and supporting their role as environmental sentinels at the human–animal–environment interface.
The second study evaluates the occurrence of antimicrobial-resistant bacteria in marine bivalves (Mytilus galloprovincialis) collected along the Ligurian coastline within the “Mare in Rete” surveillance project. The results demonstrate the presence of E. coli strains exhibiting resistance to commonly used antimicrobial classes, particularly β-lactams and tetracyclines, with higher frequencies observed in coastal areas exposed to urban and wastewater-related pressures.
Overall, the findings of this thesis highlight the relevance of wildlife and marine environments as early indicators of antimicrobial resistance circulation. The integration of terrestrial and marine monitoring provides valuable insights into environmental pathways of AMR dissemination and supports the implementation of coordinated One Health surveillance strategies at regional and national levels. Strengthening environmental monitoring systems is essential to improve risk assessment, guide preventive interventions, and mitigate the long-term impact of antimicrobial resistance on public health
Water cooled lithium lead balance of plant indirectly coupled with power conversion system operated with small energy storage in EU DEMO
No full textEuropean DEMOnstration fusion power plant (EU-DEMO) has the main goal of demonstrating the possibility of producing hundreds of MWe of electrical power from fusion by the end of the century. In this sense an important role is played by the Balance of Plant (BoP). BoP includes, at least, the Power Conversion System (PCS) that converts thermal power into electricity. PCS is thermally connected to the Primary Heat Transfer System (PHTS) which collects the thermal power generated in the reactor. In case of DEMO equipped by a Water Cooled Lithium Lead Breeding Blanket (WCLL BB), three different architectures of BoP are under development. In this work will be analyzed the Indirect BoP architecture, so called ICD (Indirect Concept Design) PCS, characterized by an Intermediate Heat Transport System (IHTS) installed between PHTS and PCS and equipped with a Small Energy Storage (molten salt) to operate, during Dwell, the Steam Turbine at around 10 % nominal steam load, as in the reference cycle. In this work it will be highlighted: i) the PCS layout and the relative Heat&Mass balance, ii) the cycle performance analysis and iii) the performance comparison with the other "pulsed" PCS architectures above mentioned. Some hints on the concept design of the interfacing Intermediate Heat Transfer Circuit with Small Energy Storage will be also provided