Abstract | ||
---|---|---|
In many applications it is desirable to learn from several kernels. \Multiple kernel learn- ing" (MKL) allows the practitioner to opti- mize over linear combinations of kernels. By enforcing sparse coecien ts, it also general- izes feature selection to kernel selection. We propose MKL for joint feature maps. This provides a convenient and principled way for MKL with multiclass problems. In addition, we can exploit the joint feature map to learn kernels on output spaces. We show the equiv- alence of several dieren t primal formulations including dieren t regularizers. We present several optimization methods, and compare a convex quadratically constrained quadratic program (QCQP) and two semi-innite linear programs (SILPs) on toy data, showing that the SILPs are faster than the QCQP. We then demonstrate the utility of our method by ap- plying the SILP to three real world datasets. |
Year | DOI | Venue |
---|---|---|
2007 | 10.1145/1273496.1273646 | ICML |
Keywords | Field | DocType |
linear combination,different regularizers,joint feature map,multiple kernel learning,generalizes feature selection,multiclass problem,different primal formulation,semi-infinite linear program,kernel selection,multiclass multiple kernel learning,convex quadratically,feature selection,quadratically constrained quadratic program,linear program | Kernel (linear algebra),Linear combination,Feature selection,Quadratically constrained quadratic program,Pattern recognition,Computer science,Multiple kernel learning,Exploit,Regular polygon,Equivalence (measure theory),Artificial intelligence,Machine learning | Conference |
Citations | PageRank | References |
113 | 5.57 | 17 |
Authors | ||
2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Alexander Zien | 1 | 1255 | 146.93 |
Cheng Soon Ong | 2 | 1232 | 86.27 |