Second-order Conditional Gradients

  • 2020-02-20 17:52:18
  • Alejandro Carderera, Sebastian Pokutta
  • 4

Abstract

Constrained second-order convex optimization algorithms are the method ofchoice when a high accuracy solution to a problem is needed, due to thequadratic convergence rates these methods enjoy when close to the optimum.These algorithms require the solution of a constrained quadratic subproblem atevery iteration. In the case where the feasible region can only be accessedefficiently through a linear optimization oracle, and computing first-orderinformation about the function, although possible, is costly, the coupling ofconstrained second-order and conditional gradient algorithms leads tocompetitive algorithms with solid theoretical guarantees and good numericalperformance.

 

Quick Read (beta)

loading the full paper ...