aalto1 untyped-item.component.html
Variational inference approaches to Bayesian optimal experimental design
Loading...
URL
Journal Title
Journal ISSN
Volume Title
Sähkötekniikan korkeakoulu |
Bachelor's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Authors
Date
Department
Major/Subject
Mcode
ELEC3056
Degree programme
Language
en
Pages
21+8
Series
Abstract
Optimal experimental design (OED) is a sub-field of statistics, that aims to optimize experiments for maximum gain of information. It is crucial for scientific research, as it allows users to design their experiments with a focus on data efficiency. Bayesian optimal experimental design (BOED) furthermore builds upon OED by incorporating prior knowledge and quantifying uncertainty into tangible variables through probability distributions. However, these probability distributions can be complex and intractable, requiring significant computational resources to calculate. This can prove infeasible in realistic situations.
To address these limitations, variational inference (VI) has proven to be a powerful tool whilst maintaining accuracy. VI replaces these intractable distributions with simpler, parametric approximations, that are obtained by minimizing either a Kullback-Leibler (KL) divergence or the evidence lower bound (ELBO). This enables efficient computation of the required distributions, making BOED feasible in high-dimensional settings.
This thesis explores the use of VI within the field of BOED, providing a thorough analysis of the state-of-the-art approaches. For this purpose, the mathematical foundations of VI are presented, including the KL divergence and the ELBO, as well as the fundamental concepts of OED and BOED. Finally, this thesis identifies the pitfalls of current approaches and discusses future research directions and potential applications aimed at making BOED more robust, accurate and scalable.