Events by date

Title: How can we optimize nonsmooth objectives globally?

Speaker: Andreas Griewank Affiliation: Humboldt University, Germany Room: MC 5501

Abstract:

In machine learning objective functions that are only piecewise smooth and should be globally minimized abound. The standard method of dealing with them is to apply a stochastic gradient method disregarding the rare points of nonsmoothness and hoping for the best as far as global optimality of the computed solution is concerned.

Title: Inverses of Trees

Speaker: Krystal Guo Affiliation: Centre de Recherches Mathématiques & Université de Montréal Room: MC 5417

Abstract:

A tree is invertible if and only if it has a perfect matching.

Godsil considers an invertible tree T and finds that the inverse of the adjacency matrix has entries in {0, ±1} and is the signed adjacency matrix of a graph which contains T.