The next decade of cosmological observations promises unprecedented datasets that will revolutionize our understanding of the Universe. However, traditional likelihood-based methods and summary statistics may prove insufficient to fully capture all available information and to account for complex theoretical and systematic uncertainties. This thesis explores simulation-based, likelihood-free inference techniques specifically designed for upcoming galaxy surveys. By maximising the mutual or Fisher information directly from simulations, these approaches can produce robust and nearly sufficient summaries of high-dimensional cosmological observables. They can further capture non-Gaussian features and complex correlations that are typically missed by traditional two-point statistics. This framework provides a robust foundation for analyzing forthcoming data from surveys like SPHEREx and Euclid, where complex systematic effects and theoretical uncertainties pose significant challenges to traditional inference methods.
Il prossimo decennio di osservazioni cosmologiche promette dataset senza precedenti che rivoluzioneranno la nostra comprensione dell'Universo. Tuttavia, i metodi tradizionali basati sulla likelihood e le statistiche a N-punti potrebbero rivelarsi insufficienti per catturare pienamente tutta l' informazione disponibile e per tenere conto delle complesse incertezze teoriche e sistematiche osservative. Questa tesi esplora tecniche di inferenza likelihood-free basate su simulazioni, specificamente progettate per le future survey galattiche. Massimizzando l'informazione mutua o di Fisher direttamente dalle simulazioni, questi approcci possono produrre sintesi robuste di osservabili cosmologici ad alta complessità. Possono inoltre catturare caratteristiche non gaussiane e correlazioni complesse che tipicamente sfuggono alle tradizionali statistiche a due punti. Questo framework fornisce una solida base per l'analisi dei futuri dati provenienti da survey come SPHEREx ed Euclid, dove effetti sistematici complessi e incertezze teoriche pongono sfide significative ai metodi di inferenza tradizionali.
Inferenza Likelihood-Free per la Nuova Generazione di Survey Galattiche: Ottimizzazione dell'Estrazione di Informazioni attraverso Machine Learning
BREGOLIN, GIACOMO
2024/2025
Abstract
The next decade of cosmological observations promises unprecedented datasets that will revolutionize our understanding of the Universe. However, traditional likelihood-based methods and summary statistics may prove insufficient to fully capture all available information and to account for complex theoretical and systematic uncertainties. This thesis explores simulation-based, likelihood-free inference techniques specifically designed for upcoming galaxy surveys. By maximising the mutual or Fisher information directly from simulations, these approaches can produce robust and nearly sufficient summaries of high-dimensional cosmological observables. They can further capture non-Gaussian features and complex correlations that are typically missed by traditional two-point statistics. This framework provides a robust foundation for analyzing forthcoming data from surveys like SPHEREx and Euclid, where complex systematic effects and theoretical uncertainties pose significant challenges to traditional inference methods.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tesi_Giacomo_Bregolin-4-1.pdf
accesso aperto
Dimensione
2.18 MB
Formato
Adobe PDF
|
2.18 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/84795