Multiple Kernel Dimensionality Reduction via Ratio-Trace and Marginal Fisher Analysis

Algorithm 1

MKL-MFA algorithm.

Input:.

Output:

Step: Initialization: ,  .

Step: Compute , and , where and are the non-zero Eigenvalue-Eigenvector

   Pairs of and respectively.

Step: Solve the SILP (19) to obtain as follows:

   While

    For

      Compute by solving =

    end

    If break;

    else

     Add to the constraint set . Update and by solving restricted Version of (19) using only

    end

    = +1;

   end

Step: Solve the ratio-trace problem(3) using spectral regression with

   To get optimal .

Step: The new non-linearly transformed representation for a data sample can be computed by

    and