Cover Image

Supernova cosmology for the 21st century : how I learnt to stop worrying about likelihoods and train a neural network /

This thesis breaks new ground in supernova type Ia cosmology, developing novel and powerful machine-learning methods scalable to the next generation of astronomical surveys. It demonstrates the feasibility of a fully simulation-based approach to inference, which overcomes the limitations of current...

Full description

Bibliographic Details
Main Author: Karchev, Konstantin (Author)
Format: eBook
Language:English
Published: Cham : Springer, [2026]
Series:Physics and Astronomy Series.
Subjects:
Supernovae as distance indicators.
Supernovae.
Electronic books.
Table of Contents:
  • Intro
  • Foreword
  • Abstract
  • Publications
  • Contents
  • Acknowledgements
  • Preface: The story so far
  • Part I Simulation-based inference
  • Chapter 1 Bayesian inference
  • 1.1 Bayesian hierarchical modelling
  • 1.2 More or less established methods
  • 1.3 The case for likelihood-free inference
  • Chapter 2 Neural simulation-based inference
  • 2.1 Flavours of neural SBI
  • 2.1.1 Neural data summarisation
  • 2.1.2 Neural density estimation
  • 2.1.3 Neural ratio estimation
  • 2.2 Inside the black box
  • 2.2.1 Augmented training
  • 2.2.2 Sequential training
  • 2.3 Verification of amortised SBI
  • 2.3.1 Training diagnostics
  • 2.3.2 Coverage tests (P-P plots)
  • Chapter 3 Neural simulation-based model selection
  • 3.1 Bayesian model selection
  • 3.2 Simulation-based model selection
  • 3.2.1 Verification of amortised model selection
  • 3.2.2 Occam's razor
  • Chapter 4 Developments in hierarchical SBI
  • 4.1 Complete hierarchical TMNRE
  • 4.2 Catalogue-based NRE
  • 4.2.1 Apologia of SBI with stochastic cardinality
  • Part II Supernova cosmology
  • Chapter 5 Supernova cosmology for philosophers
  • 5.1 A brief history of novælty
  • 5.2 A crash course in cosmologygraphy
  • 5.2.1 Confrontational theses
  • Chapter 6 Supernova cosmology for Nobel laureates
  • Chapter 7 Supernova cosmology for data scientists
  • 7.1 Digital photometry: how raw can you go
  • 7.1.1 Transient photometry: light curves and surveys
  • 7.2 The data delugsieon
  • Chapter 8 Supernova cosmology for statisticians
  • 8.1 SN Ia templates
  • 8.1.1 The de facto standard(isation)
  • 8.1.2 The Bayesian SN Ia template
  • 8.2 Bayesian SN Ia cosmology
  • 8.3 Pitfalls
  • 8.3.1 Scalability
  • 8.3.2 Redshifts (and velocities)
  • Chapter 11 SLiCsim: light curves for the ML era
  • Part IV Science
  • Chapter 12 SN Ia dust extinction with NRE .. SIDE-real applied to real data
  • 12.1 Forward modelling probabilistic SN Ia light curves
  • 12.1.1 Fake Mock data
  • 12.1.2 Real data
  • 12.2 Hierarchical NRE with the Super Tuple™
  • 12.2.1 NRE training
  • 12.2.2 Validation with HMC
  • 12.3 Results and discussion
  • 12.3.1 Comparison of marginal posteriors for the mock data
  • 12.3.2 Results on real data
  • Summary and outlook
  • Chapter 13 Simulation-based SN Ia model selection
  • 13.1 The selection of models