Deep unsupervised machine learning in the presence of missing data

Aug 2, 2026

Upload by Asher

Advances in deep statistical models have reshaped modern data-driven applications, showing strong empirical performance across many domains. Yet a substantial portion of real-world settings rely on incomplete data, which limits the practical use of deep models. This thesis tackles challenges caused by missing data for two core tasks: parameter estimation from incomplete training datasets and missing data imputation, with a focus on variational autoencoders (VAEs) and related statistical learning.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages199

Reading Time8 h 21 min

AccessFree to download

Declaration

Acknowledgements

Abstract

Missing data imputation with pre-trained VAEs

VAE estimation from incomplete training data

Variational Gibbs inference (VGI)

This thesis has been submitted in fulfilment of the requirements for a postgraduate degree (e. g. PhD, MPhil, DClinPsychol) at the University of Edinburgh. Please note the following terms and conditions of use: • This work is protected by copyright and other intellectual property rights, which are retained by the thesis author, unless otherwise stated. • A copy can be downloaded for personal non-commercial research or study, without prior permission or charge. • This thesis cannot be reproduced or quoted extensively from without first obtaining permission in writing from the author. • The content must not be changed in any way or sold commercially in any format or medium without the formal permission of the author. • When referring to this work, full bibliographic details including the author, title, awarding institution and date of the thesis must be given. Deep unsupervised machine learning in the presence of missing data Vaidotas Šimkus A thesis submitted in fulfilment of the requirements for the degree of Philosophiae Doctor in Data Science and Artificial Intelligence University of Edinburgh School of Informatics, University of Edinburgh, 10 Crichton Street, Edinburgh, EH8 9AB Vaidotas Šimkus © 2024 Declaration I declare that this thesis was composed by me, that the work contained herein is my own except where explicitly stated otherwise in the text, and that this work has not been submitted for any other degree or professional qualification except as specified. Vaidotas Šimkus November 2024 Acknowledgements I feel very fortunate to have been advised by my supervisor Michael Gutmann, whose exemplary mentorship has made my PhD journey truly rewarding. Under your guidance, I have learnt the invaluable lessons and skills that define a good researcher, including the importance of occasionally stepping back to move forward with greater clarity. Your patient mentorship was key in developing my analytical abilities and research acumen. I extend my gratitude to Chris Williams and Arno Onken, who served on my PhD committee, for their thoughtful feedback and invaluable guidance during our annual review meetings. I would also like express my thanks to the entire community within Informatics, especially Simon, Titas, Samuel, Ben, Steven, Michał, and Sandy. You have not only broadened my academic perspectives, but also inspired me to be more ambitious. The PhD journey during a global pandemic would not have been nearly as enjoyable without my wonderful friends. Mantas, Ugn , Simon, Sofija, Ivo, Titas—your stimulating and impassioned discussions recharged my mind and spirit, providing the much-needed balance to the PhD life. Edita, Indr , Tomas, Giedr , Emilis—thank you for the annual hikes and for finding time for our re-unions. Your friendship filled my journey with cherished moments that I will forever treasure. I will also be forever grateful to my family and my inspiring grandparents for the steadfast support throughout my endeavours. My deepest gratitude goes to my beloved mother, Skirmut , who sadly did not get to see this journey unfold. Yet her lasting legacy of kindness, curiosity, and belief in me was the wind at my back, propelling me forward. Finally, I would like to thank my partner, Ana, for her gentle support throughout this endeavour. Your unconditional love and kindness provided me a sanctuary that relieved the stresses and frustrations inherent to such an endeavour. Thank you, Ana, for being my pillar of strength and for gently reminding me to embrace life’s joys even when the PhD journey became overwhelming. This achievement would have been much more daunting without you by my side. Abstract Advances in deep statistical models have re-shaped modern data-driven applications, demonstrating remarkable empirical success across diverse domains. However, while some domains benefit from an abundance of clean and fully-observed data, enabling the practitioners to reap the full-be

FAQs

What two main challenges does the thesis address regarding missing data?

The thesis addresses missing data imputation and parameter estimation from incomplete training datasets.

How does the thesis approach missing data imputation with pre-trained VAEs?

It studies conditional sampling limitations in VAEs and proposes two new methods based on Markov chain Monte Carlo and importance sampling.

What issue is reported for fitting VAEs with incomplete training data?

It reports a previously unknown phenomenon where missing data hinders effective VAE fitting, and proposes variational-mixture-based strategies to mitigate it.

Aug 2, 2026

Upload by Asher