Theory Seminar, February 16, 2:00pm
Warren Weaver Hall, Room 312
Convex Geometry and Lattice Problems: New
Connections and Algorithms
Daniel Dadush,
Georgia Tech
In
this talk, I will survey newly discovered relationships and
algorithms for some fundamental lattice and convex geometry problems:
the
shortest / closest vector problem (SVP & CVP), integer programming
(IP),
and volume estimation for convex bodies. Here we obtain algorithmic
improvements from the perspectives of derandomization (SVP, CVP and
volume
estimation) and increased efficiency (IP). I will describe major
technical
tools behind our improvements, including the use of the M-ellipsoid
from convex
geometry - an ellipsoid with good covering properties w.r.t. a convex
body -
and extremal lattices from the geometry of numbers - lattices for which
the
packing and covering properties w.r.t. a convex body are nearly
identical.