TIME: 1:30 -- 2:30 pm, Tuesday, Feb 3, 2009 SPECIAL PLACE: CSE 503 SPEAKER: James Lee, University of Washington TITLE: Random embeddings, topological complexity, and the geometry of graphs ABSTRACT: It's not too difficult to show that the vertices of an n-cycle cannot be mapped into a tree without significantly distorting the pairwise distances. On the other hand, deleting a uniformly random edge of the cycle preserves every pairwise distance up to a factor of 2 in *expectation*. How general is this phenomenon? If we identify an arbitrary subset of nodes in a tree, can the resulting object still be randomly embedded into a tree? What about the same question, where "tree" is replaced by "planar graph"? In this talk, I will address (and largely resolve) the general problem of reducing "topological complexity" via random embeddings. The fact that this actually works is somewhat surprising, given that most previous results in this area had been negative. The primary application of our techniques is to a widely-studied conjecture of Gupta, Newman, Rabinovich, and Sinclair, about generalizations of the max-flow/min-cut theorem in graphs. One of our main tools arises out of a connection with the Lipschitz extension problem, and motivates our construction of probabilistic versions of some classical objects from analysis (e.g. random Whitney decompositions).