Self-organized learning of 3 dimensions

The geometric learning capabilities of a competitive neural network are studied. It is shown that the appropriate selection of a neural activity function enables the learning of the 3D geometry of a world, from two of the 2D projections of 3D extended objects.<
>