Search papers, labs, and topics across Lattice.
This paper establishes convergence guarantees for mirror descent and proximal mirror descent algorithms utilizing logarithmic barriers as distance-generating functions, addressing challenges posed by boundary solutions where Bregman divergence becomes problematic. The authors demonstrate that both algorithms achieve a convergence rate of \(O(\log k / k)\), which they prove to be tight under specific conditions. Additionally, they introduce a novel technique to manage divergence blow-up, clarify the theory of relative smoothness, and compare their approach to traditional interior-point methods.
Achieving a tight convergence rate of \(O(\log k / k\) for mirror descent algorithms could redefine optimization strategies in constrained settings.
This work derives convergence guarantees for mirror descent and proximal mirror descent algorithms when a logarithmic barrier is used as a distance-generating function. Standard approaches cannot be applied when the solution lies on the boundary, where the Bregman divergence blows up. We show that, in a specific setting, both methods enjoy an $O(\log k / k)$ rate, which is also tight. In addition, our contributions include: (i) a new technique for handling the blow-up; (ii) a resolution of a gap in the theory of relative smoothness; and (iii) a comparison of the proposed approach with interior-point methods.
