Order estimation in affine state-space neural networks

Abstract

The problem of order evaluation for an affine state-space neural network or equivalently the estimation of the number of neurons to be inserted in the hidden layer in a recurrent neural network is here addressed. The proposed method is based on a singular value decomposition applied to an oblique subspace projection given as the projection of the row space of future outputs into the past inputs-outputs row space, along the future inputs row space.