Another option could be to use an ensemble of exemplar svms [1], essentially training separate classifiers with a single target row set to 1 and all others to 0

Another option could be to use an ensemble of exemplar svms [1], essentially training separate classifiers with a single target row set to 1 and all others to 0. This can be useful when trying to pick out examples that match a very specific pattern. [1] http://www.cs.cmu.edu/~tmalisie/projects/iccv11/ http://www.cs.cmu.edu/~tmalisie/projects/iccv11/