Title:  Primal-dual first-order methods for conic optimization.

Abstract: First-order proximal methods are increasingly popular for large-scale
conic optimization, because of their low complexity per iteration compared
to interior-point methods.  The use of generalized (non-Euclidean)
distances, tailored to special problem structure, can further reduce the
complexity.  We will discuss applications in signal processing and control
that are usually handled via semidefinite programming formulations and
interior-point methods.
We will also discuss implications of self-dual structure for primal-dual
first-order methods, motivated by the importance of self-dual embedding
in modern interior-point solvers.