Main content

Analyzing Unsupervised Representation Learning Models Under the View of Dynamical Systems

Show full item record

Title: Analyzing Unsupervised Representation Learning Models Under the View of Dynamical Systems
Author: Im, Jiwoong
Department: School of Engineering
Program: Engineering
Advisor: Taylor, Graham W
Abstract: The objective of this thesis is to take the dynamical systems approach to understand the unsupervised learning models and learning algorithms. Gated auto-encoders (GAEs) are an interesting and flexible extension of auto-encoders which can learn transformations among different images or pixel covariances within images. We examine the GAEs' ability to represent different functions or data distributions. We apply a dynamical systems view to GAEs, deriving a scoring function, and drawing connections to RBMs. In the second part of our study, we investigate the performance of Minimum Probability Flow (MPF) learning for training restricted Boltzmann machines (RBMs). MPF proposes a tractable, consistent, objective function defined in terms of a Taylor expansion of the KL divergence with respect to sampling dynamics. We propose a more general form for the sampling dynamics in MPF, and explore the consequences of different choices for these dynamics for training RBMs.
URI: http://hdl.handle.net/10214/8809
Date: 2015-04
Rights: An error occurred on the license name.An error occurred on the license name.An error occurred on the license name.


Files in this item

Files Size Format View Description
Im_Jiwoong_20150505_Msc.pdf 1.147Mb PDF View/Open thesis

This item appears in the following Collection(s)

Show full item record