Work

Approximate Bayesian Computation with Sequential Surrogate Likelihoods

Year: 2019
Type: dissertation
Abstract: The purpose of this thesis was to implement, analyze, and possibly expand a Bayesian\ninference method related to approximate Bayesian computation (ABC). This method\nwas initially suggested by the supervisor and was given the working name approximate\nBayesian computation with sequential surrogate likelihoods (ABC-SSL). The\nunderlying idea for the method was to replace ABC distances with predicted distances\nobtained using some regression technique, thus circumventing generation of\nsynthetic datasets from the Bayesian model. These predictions would then be improved\nin a sequential manner, leading to a significant decrease of computational\ncost for parameter inference.\nThe literature on ABC was studied in search of similar techniques with the intent\nof finding suitable methods to be compared to ABC-SSL in a simulation study.\nGaussian process regression was chosen to model the distances due to the need for\nflexibility. An interpretation and generalization of the preliminary ABC-SSL method\nwas given, relating it to some of the methods found in the literature. The simulation\nstudy was constructed with three examples, including one of the standard models in\nthe ABC literature, the g-and-k distribution. These examples were chosen to give\nan understanding of potential use of the method. Due to lack of promising results\nof these numerical results, the complexity of the tested models were kept low.\nNo conclusive evidence was found for the inference method to be suitable for practical\nuse in its current state due to questionable asymptotic properties and difficulties\nin finding appropriate surrogate models. One possible application is to use the\nproposed technique to find regions of suspected high posterior probability of the\nparameter space to be used in combination with traditional ABC methods. Another\npossibility is to consider Bayesian optimization problems, although such problems\nwere not explicitly investigated in this thesis.\n
Author: Filip Wikman
Language: English
Open Access status: green
OpenAlex ID: W2910462821
Cites: 27
Cited by:
Related to: 10