In Bayesian statistics, a continuity property of the posterior distribution with respect to the observable variable is crucial as it expresses well-posedness, i.e., stability with respect to errors in the measurement of data. Essentially, this requires to analyze the continuity of a probability kernel or, equivalently, of a conditional probability distribution with respect to the conditioning variable. Here, we give general conditions for the Lipschitz continuity of probability kernels with respect to metric structures arising within the optimal transport framework, such as the Wasserstein metric. For dominated probability kernels over finite-dimensional spaces, we show Lipschitz continuity results with a Lipschitz constant enjoying explicit bounds in terms of Fisher-information functionals and weighted Poincar\'e constants. We also provide results for kernels with moving support, for infinite-dimensional spaces and for non dominated kernels. We show applications to several problems in Bayesian statistics, such as approximation of posterior distributions by mixtures and posterior consistency.

Lipschitz continuity of probability kernels in the optimal transport framework

Edoardo Mainini
2023-01-01

Abstract

In Bayesian statistics, a continuity property of the posterior distribution with respect to the observable variable is crucial as it expresses well-posedness, i.e., stability with respect to errors in the measurement of data. Essentially, this requires to analyze the continuity of a probability kernel or, equivalently, of a conditional probability distribution with respect to the conditioning variable. Here, we give general conditions for the Lipschitz continuity of probability kernels with respect to metric structures arising within the optimal transport framework, such as the Wasserstein metric. For dominated probability kernels over finite-dimensional spaces, we show Lipschitz continuity results with a Lipschitz constant enjoying explicit bounds in terms of Fisher-information functionals and weighted Poincar\'e constants. We also provide results for kernels with moving support, for infinite-dimensional spaces and for non dominated kernels. We show applications to several problems in Bayesian statistics, such as approximation of posterior distributions by mixtures and posterior consistency.
File in questo prodotto:
File Dimensione Formato  
23-AIHP1389-2.pdf

accesso aperto

Tipologia: Documento in versione editoriale
Dimensione 581.11 kB
Formato Adobe PDF
581.11 kB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11567/1133435
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
social impact