Training in pattern recognition from a small number of observations using projections onto null-space

Based on the algebraic approach, a system is trained in pattern recognition from a small number of observations. The feature space is built and tolerances are set up using a null-space of matrices composed of the feature vectors of the learning objects.