Georgia Institute of Technology College of Sciences

We will discuss two types of structured multi-objective optimization programs.  In the first, the goal is to minimize a sum of functions described by a graph: each function is associated with a vertex, and there is an edge between vertices if two functions share a subset of their variables.   Problems of this type arise in state estimation problems, including simultaneous localization and mapping (SLAM) in robotics, tracking, and streaming reconstruction problems in signal processing.  We will show that under mild smoothness conditions, these types of problems exhibit a type of locality: if a node is added to the graph (changing the optimization problem), the optimal solution changes only for variables that are ``close’’ to the added node, immediately giving us a quick way to update the solution as the graph grows.

In the second part of the talk, we will consider a multi-task learning problem where the solutions are expected to lie in a low-dimensional subspace.  This corresponds to a low-rank matrix recover problem where the columns of the matrix have been ``sketched’’ independently.  We show that a novel convex relaxation of this problem results in optimal sample complexity bounds.  These bounds demonstrate the statistical leverage we gain by solving the problem jointly over solving each individually.

We study reducible surgeries on knots in S^3, developing thickness bounds for L-space knots that admit reducible surgeries and lower bounds on the slice genus of general knots that admit reducible surgeries. The L-space knot thickness bounds allow us to finish off the verification of the Cabling Conjecture for thin knots. Our techniques involve the d-invariants and mapping cone formula from Heegaard Floer homology. This is joint work with Holt Bodish.

A distinctive feature of nonlinear evolution equations is the possibility of breakdown of solutions in finite time. This phenomenon, which is also called singularity formation or blowup, has both physical and mathematical significance, and, as a consequence, predicting blowup and understanding its nature is a central problem of the modern analysis of nonlinear PDEs.

In this talk we concentrate on wave maps – a geometric nonlinear wave equation – and we discuss the existence and stability of self-similar solutions, as in all higher dimensions they appear to drive the generic blowup behavior. We outline a novel framework for studying global-in-space stability of such solutions; we then men-tion some long-awaited results that we thereby obtained, and, finally, we discuss the new mathematical challenges that our approach generates.

Gaussian processes (GPs) are widely employed as versatile modeling and predictive tools in spatial statistics, functional data analysis, computer modeling and diverse applications of machine learning. They have been widely studied over Euclidean spaces, where they are specified using covariance functions or covariograms for modelling complex dependencies. There is a growing literature on GPs over Riemannian manifolds in order to develop richer and more flexible inferential frameworks. While GPs have been extensively studied for asymptotic inference on Euclidean spaces using positive definite covariograms, such results are relatively sparse on Riemannian manifolds. We undertake analogous developments for GPs constructed over compact Riemannian manifolds. Building upon the recently introduced Matérn covariograms on a compact Riemannian manifold, we employ formal notions and conditions for the equivalence of two Matérn Gaussian random measures on compact manifolds to derive the microergodic parameters and formally establish the consistency of their maximum likelihood estimates as well as asymptotic optimality of the best linear unbiased predictor.