Multi-class Discriminant Kernel Learning via Convex Programming

By Jieping Ye, Shuiwang Ji, and Jianhui Chen (submitted, 08/21/07)

 

Source codes in MATLAB:

• Binary-class case
  • SDP1 (Kernel learning by Semi-definite Programming, with regularization parameter fixed at 0.0005 in our experiments)
  • SDP2 (Joint kernel and regularization parameters learning by Semi-definite Programming)
  • QCQP1 (Kernel learning by Quadratically Constrained Quadratic programming, with regularization parameter fixed at 0.0005 in our experiments)
  • QCQP2 (Joint kernel and regularization parameters learning by Quadratically Constrained Quadratic programming)
  • SILP1 (Kernel learning by Semi-infinite Linear Programming with regularization parameter fixed at 0.0005 and tolerance parameter fixed at 0.0005 in our experiments)
  • SILP2 (Joint kernel and regularization parameters learning by Semi-infinite Linear Programming)
• Multi-class case
  • SDP1  (Kernel learning by Semi-definite Programming, with regularization parameter fixed at 0.0005 in our experiments)
  • SDP2 (Joint kernel and regularization parameters learning by Semi-definite Programming)
  • QCQP1 (Kernel learning by Quadratically Constrained Quadratic programming, with regularization parameter fixed at 0.0005 in our experiments)
  • QCQP2 (Joint kernel and regularization parameters learning by Quadratically Constrained Quadratic programming)
  • SILP1 (Kernel learning by Semi-infinite Linear Programming with regularization parameter fixed at 0.0005 and tolerance parameter fixed at 0.0005 in our experiments)
  • SILP2 (Joint kernel and regularization parameters learning by Semi-infinite Linear Programming)

   

Link to SeDuMi package

Link to MOSEK package