Particle-Based Online Bayesian Learning of Static Parameters with Application to Mixture Models

This thesis investigates the possibility of using Sequential Monte Carlo methods (SMC) to create an online algorithm to infer properties from a dataset, such as unknown model parameters. Statistical inference from data streams tends to be difficult, and this is particularly the case for parametric m...

Full description

Bibliographic Details
Main Author: Fuglesang, Rutger
Format: Others
Language:English
Published: KTH, Matematisk statistik 2020
Subjects:
SMC
Online Access:http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-279847
id ndltd-UPSALLA1-oai-DiVA.org-kth-279847
record_format oai_dc
spelling ndltd-UPSALLA1-oai-DiVA.org-kth-2798472020-09-05T05:28:00ZParticle-Based Online Bayesian Learning of Static Parameters with Application to Mixture ModelsengPartikelbaserad Bayesiansk realtidsinlärning av statiska modellparameterar med tillämpning på mixturmodellerFuglesang, RutgerKTH, Matematisk statistik2020Statisticsapplied mathematicssequential monte carloSMCStatistical inferenceStatisksekventiell monte carloSMCtillämpad matematikProbability Theory and StatisticsSannolikhetsteori och statistikThis thesis investigates the possibility of using Sequential Monte Carlo methods (SMC) to create an online algorithm to infer properties from a dataset, such as unknown model parameters. Statistical inference from data streams tends to be difficult, and this is particularly the case for parametric models, which will be the focus of this paper. We develop a sequential Monte Carlo algorithm sampling sequentially from the model's posterior distributions. As a key ingredient of this approach, unknown static parameters are jittered towards the shrinking support of the posterior on the basis of an artificial Markovian dynamic allowing for correct pseudo-marginalisation of the target distributions. We then test the algorithm on a simple Gaussian model, a Gausian Mixture Model (GMM), as well as a variable dimension GMM. All tests and coding were done using Matlab. The outcome of the simulation is promising, but more extensive comparisons to other online algorithms for static parameter models are needed to really gauge the computational efficiency of the developed algorithm. Detta examensarbete undersöker möjligheten att använda Sekventiella Monte Carlo metoder (SMC) för att utveckla en algoritm med syfte att utvinna parametrar i realtid givet en okänd modell. Då statistisk slutledning från dataströmmar medför svårigheter, särskilt i parameter-modeller, kommer arbetets fokus ligga i utvecklandet av en Monte Carlo algoritm vars uppgift är att sekvensiellt nyttja modellens posteriori fördelningar. Resultatet är att okända, statistiska parametrar kommer att förflyttas mot det krympande stödet av posterioren med hjälp utav en artificiell Markov dynamik, vilket tillåter en korrekt pseudo-marginalisering utav mål-distributionen. Algoritmen kommer sedan att testas på en enkel Gaussisk-modell, en Gaussisk mixturmodell (GMM) och till sist en GMM vars dimension är okänd. Kodningen i detta projekt har utförts i Matlab. Student thesisinfo:eu-repo/semantics/bachelorThesistexthttp://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-279847TRITA-SCI-GRU ; 2020:306application/pdfinfo:eu-repo/semantics/openAccess
collection NDLTD
language English
format Others
sources NDLTD
topic Statistics
applied mathematics
sequential monte carlo
SMC
Statistical inference
Statisk
sekventiell monte carlo
SMC
tillämpad matematik
Probability Theory and Statistics
Sannolikhetsteori och statistik
spellingShingle Statistics
applied mathematics
sequential monte carlo
SMC
Statistical inference
Statisk
sekventiell monte carlo
SMC
tillämpad matematik
Probability Theory and Statistics
Sannolikhetsteori och statistik
Fuglesang, Rutger
Particle-Based Online Bayesian Learning of Static Parameters with Application to Mixture Models
description This thesis investigates the possibility of using Sequential Monte Carlo methods (SMC) to create an online algorithm to infer properties from a dataset, such as unknown model parameters. Statistical inference from data streams tends to be difficult, and this is particularly the case for parametric models, which will be the focus of this paper. We develop a sequential Monte Carlo algorithm sampling sequentially from the model's posterior distributions. As a key ingredient of this approach, unknown static parameters are jittered towards the shrinking support of the posterior on the basis of an artificial Markovian dynamic allowing for correct pseudo-marginalisation of the target distributions. We then test the algorithm on a simple Gaussian model, a Gausian Mixture Model (GMM), as well as a variable dimension GMM. All tests and coding were done using Matlab. The outcome of the simulation is promising, but more extensive comparisons to other online algorithms for static parameter models are needed to really gauge the computational efficiency of the developed algorithm. === Detta examensarbete undersöker möjligheten att använda Sekventiella Monte Carlo metoder (SMC) för att utveckla en algoritm med syfte att utvinna parametrar i realtid givet en okänd modell. Då statistisk slutledning från dataströmmar medför svårigheter, särskilt i parameter-modeller, kommer arbetets fokus ligga i utvecklandet av en Monte Carlo algoritm vars uppgift är att sekvensiellt nyttja modellens posteriori fördelningar. Resultatet är att okända, statistiska parametrar kommer att förflyttas mot det krympande stödet av posterioren med hjälp utav en artificiell Markov dynamik, vilket tillåter en korrekt pseudo-marginalisering utav mål-distributionen. Algoritmen kommer sedan att testas på en enkel Gaussisk-modell, en Gaussisk mixturmodell (GMM) och till sist en GMM vars dimension är okänd. Kodningen i detta projekt har utförts i Matlab.
author Fuglesang, Rutger
author_facet Fuglesang, Rutger
author_sort Fuglesang, Rutger
title Particle-Based Online Bayesian Learning of Static Parameters with Application to Mixture Models
title_short Particle-Based Online Bayesian Learning of Static Parameters with Application to Mixture Models
title_full Particle-Based Online Bayesian Learning of Static Parameters with Application to Mixture Models
title_fullStr Particle-Based Online Bayesian Learning of Static Parameters with Application to Mixture Models
title_full_unstemmed Particle-Based Online Bayesian Learning of Static Parameters with Application to Mixture Models
title_sort particle-based online bayesian learning of static parameters with application to mixture models
publisher KTH, Matematisk statistik
publishDate 2020
url http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-279847
work_keys_str_mv AT fuglesangrutger particlebasedonlinebayesianlearningofstaticparameterswithapplicationtomixturemodels
AT fuglesangrutger partikelbaseradbayesianskrealtidsinlarningavstatiskamodellparameterarmedtillampningpamixturmodeller
_version_ 1719339443772456960