Classification with Sums of Separable Functions

We present a novel approach for classification using a discretised function representation which is independent of the data locations. We construct the classifier as a sum of separable functions, extending the paradigm of separated representations. Such a representation can also be viewed as a low rank tensor product approximation. The central learning algorithm is linear in both the number of data points and the number of variables, and thus is suitable for large data sets in high dimensions. We show that our method achieves competitive results on several benchmark data sets which gives evidence for the utility of these representations.