Nina Thiessen's M.Sc. Thesis

Completed September 2004
Supervisor: Geoff Hinton

Connection Maps: A New Way to Visualize Similarity Relationships

Abstract: Hinton diagrams a visualization tool that has proven helpful for visualizing the patterns of connection strengths in artificial neural networks. This thesis explores whether a tool derived from Hinton diagrams could be helpful for visualizing similarity relationships. A connection map diagram arranges the elements in a data set on a 2-dimensional grid such that similar elements are near each other. Each element is represented by its `connection map', which is an icon composed of colored squares indicating the pairwise similarity relationships of that element to all the other elements. Each colored square in a connection map is a pointer to the element at that position on the grid. Largely similar maps can be located and compared using the relative positions of differently colored squares and the overall position of this colored pattern within the connection map, a task to which our visual systems appear to be very well adapted.

Download: finalThesis2sided.ps or finalThesis2sided.pdf

Also available is some of the Matlab code used to generate the connection map diagrams as well as some of the pictures from my thesis.