1,720,955 research outputs found
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
eXplainable Checker of Video Analytics Performance in Indoor Smart Mobility
No full textEnsuring the performance of object detection systems in dynamic environments requires not only accurate predictions but also the ability to assess the certainty of those predictions. This paper proposes an interpretable framework for monitoring the Operational Design Domain (ODD) of real-time object detectors through visual feature-based certainty evaluation. Using a dual-path architecture, the system combines a standard object detection pipeline with a parallel branch that extracts visual features and classifier predictions as Certain or Uncertain using decision rules learned via Decision Trees (DTs). A Cumulative Feature Ranking (CFR) strategy ensures robust selection of discriminative features across perturbed and real-world datasets. Extensive experiments on the Pedestrian and Wheelchair object categories demonstrate the system’s ability to detect prediction uncertainty under a variety of visual conditions. The interpretable nature of the learned rules provides transparency, while the low false alarm rate demonstrates the effectiveness of the ODD checker in supporting safe and explainable perception for indoor smart mobility applications
Going Beyond Counting First Authors in Author Co-citation Analysis
Full text linkThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
Variations on the Author
Full text link“Variations on the Author” discusses two of Eduardo Coutinho’s recent films (Um Dia na Vida, from 2010, and Últimas Conversas, posthumously released in 2015) and their contribution to the general question of documentary authorship. The director’s filmography is characterized by a consistent yet self-effacing form of authorial self-inscription: Coutinho often features as an interviewer that rather than express opinions propels discourses; an interviewer that is good at listening. This mode of self-inscription characterizes him as an author who is not expressive but who is nonetheless markedly present on the screen. In Um Dia na Vida, however, Coutinho is completely absent form the image, while Últimas Conversas, on the contrary, includes a confessional prologue that moves the director from the margins to the center of his films. This article examines the ways in which these works stand out in the filmography of a director who offers new insights into the notion of cinematic authorship
Appropriate Similarity Measures for Author Cocitation Analysis
Full text linkWe provide a number of new insights into the methodological discussion about author cocitation analysis. We first argue that the use of the Pearson correlation for measuring the similarity between authors’ cocitation profiles is not very satisfactory. We then discuss what kind of similarity measures may be used as an alternative to the Pearson correlation. We consider three similarity measures in particular. One is the well-known cosine. The other two similarity measures have not been used before in the bibliometric literature. Finally, we show by means of an example that our findings have a high practical relevance.information science;Pearson correlation;cosine;similarity measure;author cocitation analysis
Dispelling the Myths Behind First-author Citation Counts
Full text linkWe conducted a full-scale evaluative citation analysis study of scholars in the XML research field to explore just how different from each other author rankings resulting from different citation counting methods actually are, and to demonstrate the capability of emerging data and tools on the Web in supporting more realistic citation counting methods. Our results contest some common arguments for the continued
use of first-author citation counts in the evaluation of scholars, such as high correlations between author rankings by first-author citation counts and other citation
counting methods, and high costs of using more realistic citation counting methods that are not well-supported by the ISI databases. It is argued that increasingly available digital full text research papers make it possible for citation analysis studies to go beyond what the ISI databases have directly supported and to employ more
sophisticated methods
koamabayili/VECTRON-author-checklist: VECTRON author checklist
No full textWe have done our best to complete the author checklist relating to the use of animals in the hut study. Note that the objective for the hut study was to evaluate the IRS treatment applications for residual efficacy against Anopheles mosquitoes, including the local An. coluzzii mosquito population. Cows were only used to attract mosquitoes into the huts and no tests were carried out directly on the cows. The author checklist is intended for use with studies where experiments are carried out on animals, which is why we have had such difficulty in completing this for the hut study, as many of the questions do not relate to how the cows were used
Author-wise bibliometric analysis based on entropy.
No full textAuthor-wise bibliometric analysis based on entropy.</p
- …
