Faculty Host: Tina Eliassi-Rad
Abstract: Graph theory can be a very powerful tool for a variety of different problems, but it isn't always clear how the edges of the graph should be defined. Link prediction is the process of determining which pairs of nodes should be connected by an edge. This seminar describes the process of developing different link prediction algorithms. We first discuss how to choose intermediary metrics that correspond well to the application of interest in order to obtain measures of performance before it is practical to obtain a measure of effectiveness for that application. We also describe the use of Lincoln's VizLinc Audio Visual tool in using visual analytics to gain better insight into the algorithms than numeric metrics alone can provide. Throughout the talk, we will use speaker recognition as the domain of interest, generating speaker content graphs that efficiently model the underlying manifold of the speaker space, and using those graphs to perform tasks such as query by example and speaker clustering.
This is joint work with Dr. William M. Campbell at MIT Lincoln Laboratory.
This work is sponsored by the Department of Defense under Air Force Contract FA8721-05-C-0002. Opinions, interpretations, conclusions, and recommendations are those of the author and are not necessarily endorsed by the United States Government.
Bio: Kara Greenfield's bio is available here.
Suggested Readings: