Registered Data

[00213] Advances in Derivative-free Methods and the DFO VU-algorithm

  • Session Time & Room : 4D (Aug.24, 15:30-17:10) @D408
  • Type : Contributed Talk
  • Abstract : The VU-algorithm is a method of minimizing convex, nonsmooth functions by splitting the space into two subspaces: the V-space, on which the objective function's nonsmooth behavior is captured, and the orthogonal U-space, on which the function behaves smoothly. The algorithm's convergence is accelerated, as it takes a (slow) proximal point step in the V-space, then a (fast) quasi-Newton step in the U-space, since gradients and Hessians exist there. New convergence rates and subroutines are presented.
  • Classification : 90C25, 49J52
  • Format : Talk at Waseda University
  • Author(s) :
    • Chayne Planiden (University of Wollongong)