SpicyMKL

SpicyMKL is an optimization method of MKL (mutiple kernel learning) that scales well against the number of kernels. The software affords hinge, logistic and square losses, and L1-norm and elasticnet regularization.

Matlab Implementation: SpicyMKL_ver3.zip (updated 1st Apr. 2014)

SpicyMKL.m is the main function.
- SpicyMKL.m calls normKj.mex and HessAugMexbias.mex. Please link blas library when compiling these files like
```
    mex -lmwblas normKj.c
    mex -lmwblas HessAugMexbias.c
```
(startup.m contains this code).
demo.m is a demo script.
- The demo code requires SimpleMKL Toolbox in the path.
If you have troubles to compile the mex-files, please use m-files contained in the folder "auxiliary_m_file". Just put the m-files into the same directory as "SpicyMKL.m".

Examples

ver2: SpicyMKL.zip (updated 1st Nov. 2010)

References

Suzuki, T. & Tomioka, R: SpicyMKL. arXiv:0909.5026,