PATTERN-CLASSIFICATION USING THE PHASE-VELOCITY FIELD APPROACH

A new concept for unsupervised learning based upon examples introduced to the neural network is proposed. Each example is considered as an interpolation node of the velocity field in the phase space. The velocities at these nodes are selected such that all the streamlines converge to an attracting set embedded in the subspace occupied by the cluster of examples. The unknown parameters are found from a learning procedure using the selected phase velocity field as an input. The theory is further ilucidated by an example.