Performance improvement of the BSB model in classifier tasks using optimization techniques

A method to improve the performance of the BSB (Brain State in a Box) model applied to classification tasks is presented. We propose a supervised learning scheme divided in two stages. In the first one the basic interconnection matrix is formed using data of a set of training points, as proposed in the original definition of the BSB model. In the second stage the behavior of the trajectories that begin in each one of these points are studied and modified by means of a change in the interconnection matrix. The correction is achieved using of a method derived from the lineal optimal control techniques, applied on the evolutions that begin in the training set points. Several simulated experiments were carried out and the effectiveness of the proposed method is shown