Classification in multi- observational setting using latent Gaussian Processes

Loading...
Thumbnail Image
Journal Title
Journal ISSN
Volume Title
Perustieteiden korkeakoulu | Master's thesis
Date
2017-12-11
Department
Major/Subject
Machine Learning and Data Mining
Mcode
SCI3044
Degree programme
Master’s Programme in Computer, Communication and Information Sciences
Language
en
Pages
46+6
Series
Abstract
Widespread interest in the usage of data collection devices all around the world has resulted in an increasingly large number of sequential multivariate datasets. Be it IoT applications, wearable sensors, medical records or fMRI records number of datasets with series of multiple observations per sample is growing. Most of these datasets typically constitute observations of a fairly complex process and contain thousands of data points. High dimensionality of these datasets combined with their susceptibility to missing data and multi observational setting can make implementing traditional data analysis techniques for these datasets challenging. Impressed with their ability to propagate prior information about latent processes and learn the components nonparametrically, we explore Bayesian latent variable models and propose a multi-observational sparse Gaussian process based classifier that can efficiently classify observations by learning separate latent space representation for each observation. As a precursor to the development of our proposed model we derived a scalable variational approximation for the semiparametric latent factor model and further extended it to accommodate multi-observational datasets. Finally, we perform several experiments and demonstrations with artificial datasets on the proposed model to ensure that model is not overly sensitive to the variability of parameters and can achieve classification performance at-par with other popular classification methods.
Description
Supervisor
Kaski, Samuel
Thesis advisor
Remes, Sami
Keywords
gaussian process, latent variable model, classification, multi-observational setting, latent gaussian process
Other note
Citation