A Unified Framework for Bregman Proximal Methods
Date:
Locations:
- INFORMS Annual Meeting 2019, Seattle, WA
In this talk we propose a novel framework for the convergence analysis of Bregman proximal methods. This framework enables us to derive the standard $\mathcal{O}(1/k) $rate for the proximal gradient method and new rates for its accelerated variant. The crux of our framework is an upper bound constructed via the convex conjugate of the objective function.