Convex transduction with the normalized cut
published: Feb. 25, 2007, recorded: March 2005, views: 3457
Report a problem or upload filesIf you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.
We discuss approaches to transduction based on graph cut cost functions. More specifically, we focus on the normalized cut, which is the cost function of choice in many clustering applications, notably in image segmentation. Since optimizing the normalized cut cost is an NP-complete problem, much of the research attention so far has gone to relaxing the problem of normalized cut clustering to tractable problems, producing so far a spectral relaxation and a more recently a tighter but computationally much tougher semi-definite programming (SDP) relaxation. In this paper we deliver two main contributions: first, we show how an alternative SDP relaxation yields a much more tractable optimization problem, and we show how scalability and speed can further be increased by making a principled approximation. Second, we show how it is possible to efficiently optimize the normalized cut cost in a transduction setting using our newly proposed approaches. Positive empirical results are reported.
Link this pageWould you like to put a link to this lecture on your homepage?
Go ahead! Copy the HTML snippet !