Multiclass Learning at One-Class Complexity
Sandor Szedmak and John Shawe-Taylor
School of Electronics and Computer Science, Southampton, UK.
We show in this paper the multiclass classification problem can be implemented in the maximum margin framework with the complexity
of one binary Support Vector Machine. We show reducing the
complexity does not involve diminishing performance but in some cases this approach can improve the classification accuracy. The multiclass
classification is realized in the framework where the output labels are vector valued.