Event time:
Tuesday, November 15, 2016 - 11:15am to 12:15pm
Location:
AKW 200
Speaker:
Nisheeth Vishnoi
Speaker affiliation:
École Polytechnique Fédérale de Lausanne
Event description:
We present an algorithmic connection between nature and humans arising in entirely different contexts. The first is the famous Iteratively Reweighted Least Squares (IRLS) algorithm used for the sparse recovery problem while the second is the dynamics of a slime mold. Despite its simplicity the convergence of the IRLS method has been shown only for a certain regularization of it and remains an important open problem. We show that the two dynamics are projections of the same dynamical system in higher dimensions. Subsequently, we take inspiration from the analysis of the slime mold dynamics to show convergence and obtain complexity bounds for a damped version of the IRLS
algorithm. Joint work with Damian Straszak.