Ranking Support Vector Machine with Kernel Approximation

Algorithm 3

RankSVM with kernel approximation.

Require: , , ,

Ensure:

  (1) Calculate the approximation embedding using the Nyström method or random Fourier features;

  (2) Apply to training samples, ;

  (3) repeat

  (4)  ;

  (5)  ;

  (6)  // Solve by linear CG

  (7)  repeat

  (8)   Update based on the computation of Hessian-vector multiplication, , for some vector ;

  (9)  until Convergence

 (10)  ;

 (11) until Convergence of the Newton step