The increasing availability of large-scale geospatial and spatiotemporal data presents new opportunities and challenges for statistical modeling in environmental, technological, medical, and other complex areas, which increasingly rely on massive multivariate spatiotemporal datasets. Yet, Bayesian learning for such problems remains severely limited by computational bottlenecks and the lack of flexible modeling tools. Modern applications require methods that are adaptive and effective, but still computationally efficient, scalable to massive datasets, and capable of delivering reliable automated inference with principled uncertainty quantification and (possibly) minimal experienced human intervention. Classical Bayesian approaches, although theoretically appealing and offering rich inferential frameworks, often become computationally infeasible in data-rich environments, especially when confronted with massive datasets or dynamic, high-dimensional dependence structures. Existing approaches often fail to scale, leaving a gap between the theoretical richness of Bayesian inference and its practical deployment in data-rich applications. This thesis develops Bayesian transfer learning methodologies to address these challenges, enabling efficient information propagation and scalable inference across large spatial and spatiotemporal domains, providing a unified framework that merges distributional theory for matrix-variate models with computational innovations in Bayesian predictive stacking. Through extensive simulation experiments and data applications to global and satellite monitoring of vegetation indices, sea surface temperature, and land-atmospheric climate composition, the thesis also demonstrates the potential of Bayesian transfer learning to redefine spatial and spatiotemporal multivariate modeling, providing flexible, computationally efficient solutions that open the way for scalable, automated, and truly modern tools for geospatial learning in data-rich environments.
La crescente disponibilità di dati geospaziali e spaziotemporali su larga scala offre nuove opportunità e, al contempo, impone sfide significative per la modellazione statistica in ambito ambientale, tecnologico, medico e in numerosi altri settori complessi, che fanno sempre più affidamento su dataset spaziotemporali multivariati di dimensioni massive. Tuttavia, l’apprendimento Bayesiano per tali problemi risulta tuttora gravemente limitato da colli di bottiglia computazionali e dalla mancanza di strumenti modellistici sufficientemente flessibili. Le applicazioni moderne richiedono metodi adattivi ed efficaci, ma al tempo stesso computazionalmente efficienti, scalabili a dataset di grandi dimensioni e in grado di fornire inferenza automatizzata e affidabile, con quantificazione dell’incertezza rigorosa e, possibilmente, un intervento umano esperto minimo. Gli approcci Bayesiani classici, pur essendo teoricamente solidi e caratterizzati da un ricco potenziale inferenziale, diventano spesso computazionalmente impraticabili in contesti con abbondanza di dati, soprattutto in presenza di dataset massivi o strutture di dipendenza dinamiche e ad alta dimensionalità. Le metodologie esistenti, infatti, raramente riescono a scalare in tali scenari, lasciando così aperto un divario tra la ricchezza teorica dell’inferenza Bayesiana e la sua effettiva applicabilità in problemi reali guidati dai dati. Questa tesi sviluppa metodologie di trasferimento dell'apprendimento Bayesiano per affrontare tali sfide, abilitando una propagazione dell’informazione efficiente e un’inferenza scalabile in domini spaziali e spaziotemporali di grande estensione, e fornendo un quadro unificato che integra la teoria distribuzionale dei modelli matriciali-variati con innovazioni computazionali basate sullo stacking predittivo Bayesiano. Attraverso ampie simulazioni e applicazioni a dati riguardanti monitoraggi globali e satellitari di indici di vegetazione, temperatura superficiale marina e composizione climatica terra–atmosfera, la tesi dimostra il potenziale del trasferimento dell'apprendimento Bayesiano nel ridefinire la modellazione spaziale e spaziotemporale multivariata, fornendo soluzioni flessibili e computazionalmente efficienti, aprendo la strada a strumenti moderni, scalabili e automatizzati per l’analisi geospaziale in ambienti caratterizzati da abbondanza di dati.
Bayesian Transfer Learning Approaches for Large-scale Spatiotemporal Problems
PRESICCE, LUCA
2026
Abstract
The increasing availability of large-scale geospatial and spatiotemporal data presents new opportunities and challenges for statistical modeling in environmental, technological, medical, and other complex areas, which increasingly rely on massive multivariate spatiotemporal datasets. Yet, Bayesian learning for such problems remains severely limited by computational bottlenecks and the lack of flexible modeling tools. Modern applications require methods that are adaptive and effective, but still computationally efficient, scalable to massive datasets, and capable of delivering reliable automated inference with principled uncertainty quantification and (possibly) minimal experienced human intervention. Classical Bayesian approaches, although theoretically appealing and offering rich inferential frameworks, often become computationally infeasible in data-rich environments, especially when confronted with massive datasets or dynamic, high-dimensional dependence structures. Existing approaches often fail to scale, leaving a gap between the theoretical richness of Bayesian inference and its practical deployment in data-rich applications. This thesis develops Bayesian transfer learning methodologies to address these challenges, enabling efficient information propagation and scalable inference across large spatial and spatiotemporal domains, providing a unified framework that merges distributional theory for matrix-variate models with computational innovations in Bayesian predictive stacking. Through extensive simulation experiments and data applications to global and satellite monitoring of vegetation indices, sea surface temperature, and land-atmospheric climate composition, the thesis also demonstrates the potential of Bayesian transfer learning to redefine spatial and spatiotemporal multivariate modeling, providing flexible, computationally efficient solutions that open the way for scalable, automated, and truly modern tools for geospatial learning in data-rich environments.| File | Dimensione | Formato | |
|---|---|---|---|
|
phd_unimib_862302.pdf
accesso aperto
Licenza:
Tutti i diritti riservati
Dimensione
8.37 MB
Formato
Adobe PDF
|
8.37 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/359626
URN:NBN:IT:UNIMIB-359626