Large databases recognition tasks: a proposal for partitioning the data matrix required to train a radial basis functions network

When a radial basis functions (RBF) network is used to perform recognition tasks, a matrix is built that contains the projections of the input vectors into the space of RBF; the dimension of this matrix depends on the number of the RBF used and on the number of vectors in the training set, i.e. the number of vectors chosen in the input space. In this paper we deal with problems arising when this number is very large, thus making it difficult for every operation we want to perform with the matrix: we suggest a technique to paginate the matrices involved in the calculations and so to obtain the result quicker.