Functional Bayesian Neural Networks with Non-stationary Gaussian Process Priors and Belief Matching

Thumbnail Image

Files

master_Papatheodorou_Dimitrios_2024.pdf (692.15 KB)

URL

Journal Title

Journal ISSN

Volume Title

Perustieteiden korkeakoulu | Master'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

2024-01-22

Department

Major/Subject

Machine Learning, Data Science and Artificial Intelligence

Mcode

SCI3044

Degree programme

Master’s Programme in Computer, Communication and Information Sciences

Language

en

Pages

59

Series

Abstract

In the realm of machine learning, overfitting remains a persistent challenge when striving for optimal performance in various tasks, and modeling uncertainty for better explainability and regularisation a driving force for more research in bayesian approaches. This master's thesis delves into the fusion of diverse concepts from Bayesian statistics, Gaussian processes, and neural networks to address the complex issues of overfitting and uncertainty quantification while maintaining accuracy. Specifically, it explores the incorporation of non-stationary kernels within functional Bayesian neural networks with Gaussian process priors, and harnessing the power of bayesian belief matching in classification to strike a balance between predictive capability and generalization. Experimental results validate the efficacy of the proposed approach, showcasing the potential of synergizing bayesian approaches with neural networks, but also highlight its limitations and complexities.

Description

Supervisor

Lähdesmäki, Harri

Thesis advisor

Lähdesmäki, Harri

Keywords

machine learning, neural networks, bayesian inference, gaussian processes

Other note

Citation

Permanent link to this item

https://urn.fi/URN:NBN:fi:aalto-202401282051

Collections

[dipl] Perustieteiden korkeakoulu / SCI