Smoothness on graphs, total variation, and clustering
Time and place
1 PM on Thursday, February 25th, 2010; NAC 6/113
Prof. Arthur Szlam (Courant Institute, NYU)
Abstract
In this talk I will start by reviewing how the relationships between smoothness, scale, and frequency that form the core of harmonic analysis persist in the setting of weighted graphs. I will illustrate these ideas with examples from image and audio processing, and then with examples from data clustering. Finally, I will discuss some of the theoretical and algorithmic consequences of replacing $\mathcal{L}^2$ notions of smoothness by $\mathcal{L}^1$ notions of smoothness in the context of the clustering problem.