[BANANA] Two LA/Opt seminars this week (Wed: Elizabeth Wong), Fri: Ya-xiang Yuan)

[Michael A. Saunders]

Dear Linear Algebra & Optimization colleagues,

We have two seminars this week (in different rooms!).
Both are about large-scale QP:

    4:15pm Wed Oct 7 Terman 332: Elizabeth Wong, UCSD
    4:15pm Fri Oct 9 Terman 453: Ya-xiang Yuan, Beijing

Next week is the INFORMS Annual Meeting in San Diego (probably no seminar).

---------------------------------------------------------------------------
    Linear Algebra and Optimization Seminar (CME 510)
    iCME, Stanford University
    http://icme.stanford.edu/seminars/seminars.php

    4:15pm Wed Oct 7, 2009
    Terman 332
    http://campus-map.stanford.edu/index.cfm?ID=01-240

    Elizabeth Wong
    Dept of Mathematics
    UC San Diego, La Jolla, CA
    http://ccom.ucsd.edu/~elwong/

A regularized method for general quadratic programming

We consider a quadratic programming method designed for use in a
sequential quadratic programming (SQP) method for large-scale
nonlinearly constrained optimization.

Because the efficiency of SQP methods is determined by how the
quadratic subproblem is formulated and solved, we propose an
active-set method based on inertia control that prevents
singularity in the associated KKT systems.  The method is able to
utilize black-box linear algebra software, thereby exploiting
recent advances in computer hardware.  Moreover, the method makes
no assumptions on the convexity of the quadratic problems, making
it particularly useful in SQP methods using exact second
derivatives.

In addition, the method can be applied to a regularized quadratic
subproblem involving an augmented Lagrangian objective function,
eliminating the need for a full-rank assumption on the constraint
matrix.


---------------------------------------------------------------------------
   SPECIAL JOINT OPTIMIZATION SEMINAR

   CME 510 Fall 2009 Linear Algebra and Optimization Seminar
   iCME, Stanford University
   http://icme.stanford.edu/seminars/seminars.php

   Operations Research @ Stanford
   http://or.stanford.edu/oras_seminars.html

   4:15pm Friday Oct 9, 2009
   Terman 453

   Professor Ya-xiang Yuan
   Institute of Computational Mathematics and Scientific/Engineering Computing
   Chinese Academy of Sciences, Beijing
   yyx <yyx at lsec.cc.ac.cn>
   http://lsec.cc.ac.cn/~yyx/

   A Parallel Decomposition Algorithm for Training
   Multiclass Kernel-based Vector Machines

We present a decomposition method for training Crammer and Singer's
multiclass kernel-based vector machine model.  A new working set
selection rule is proposed.  Global convergence of the algorithm
based on this selection rule is established.  A projected gradient
method is chosen to solve the resulting quadratic subproblem at each
iteration.  An efficient projection algorithm is designed by
exploiting the structure of the constraints.  Parallel strategies
are given to utilize the storage and computational resources
available on multiprocessor systems.  Numerical experiments on
benchmark problems demonstrate that good classification accuracy
and remarkable time saving can be achieved.

---------------------------------------------------------------------------