Georgia Institute of Technology College of Sciences

I will define planted distributions of random structures and give plenty of examples in different contexts: from balls and bins, to random permutations, to random graphs and CSP's. I will give an idea of how they are used and why they are interesting. Then I'll focus on one particular problem: under what conditions can you distinguish a planted distribution from the standard distribution on a random structure and how can you do it?

A fundamental result in the geometry of numbers states that any lattice free convex set in R^n has integer width bounded by a function of dimension, i.e. the so called Flatness Theorem for Convex Bodies. This result provides the theoretical basis for the polynomial solvability of Integer Programs with a fixed number of (general) integer variables. In this work, we provide a simplified proof of the Flatness Theorem with tighter constants. Our main technical contribution is a new tight bound on the smoothing parameter of a lattice, a concept developed within lattice based cryptography which enables comparisons between certain discrete distributions over integer points with associated continuous Gaussian distributions. Based on joint work with Kai-Min Chung, Feng Hao Liu, and Christopher Peikert.

Billiards is a dynamical system generated by an uniform motion of a point particle (ray of light, sound, etc.) in a domain with piecewise smooth boundary. Upon reaching the boundary the particle reflected according to the law "the angle of incidence equals the angle of reflection". Billiards appear as natural models in various branches of physics. More recently this type of models were used in oceanography, operations research, computer science, etc. I'll explain on very simple examples what is a regular and what is chaotic dynamics, the mechanisms of chaos and natural measures of complexity in dynamical systems. The talk will be accessible to undergraduates.

Santosh Vempala and I have been exploring an intriguing newapproach to convex optimization. Intuition about first-order interiorpoint methods tells us that a main impediment to quickly finding aninside track to optimal is that a convex body's boundary can get inone's way in so many directions from so many places. If the surface ofa convex body is made to be perfectly reflecting then from everyinterior vantage point it essentially disappears. Wondering about whatthis might mean for designing a new type of first-order interior pointmethod, a preliminary analysis offers a surprising and suggestiveresult. Scale a convex body a sufficient amount in the direction ofoptimization. Mirror its surface and look directly upwards fromanywhere. Then, in the distance, you will see a point that is as closeas desired to optimal. We wouldn't recommend a direct implementation,since it doesn't work in practice. However, by trial and error we havedeveloped a new algorithm for convex optimization, which we arecalling Reflex. Reflex alternates greedy random reflecting steps, thatcan get stuck in narrow reflecting corridors, with simply-biasedrandom reflecting steps that escape. We have early experimentalexperience using a first implementation of Reflex, implemented inMatlab, solving LP's (can be faster than Matlab's linprog), SDP's(dense with several thousand variables), quadratic cone problems, andsome standard NETLIB problems.