Abstract: Convolutional networks (12) can be defined by a set of layers (1, 2, 3) being respectively made up by a two-dimensional lattice of neurons (4). Each layer (1, 2, 3)—with the exception of the last layer (3)—represents a source layer (7) for respectively following target layer (8). A plurality of neurons (4) of a source layer (7) called a source sub-area respectively share the identical connectivity weight matrix type. Each connectivity weight matrix type is represented by a scalar product of an encoding filter (10) and a decoding filter (11). For each source layer (7) a source reconstruction image (9) is calculated on the basis of the corresponding encoding filters (10) and the activities of the corresponding source sub-area. For each connectivity weight matrix type, each target sub-area and each target layer (8) the input of the target layer (8) is calculated (S6) as a convolution of the source reconstruction image (9) and the decoding filter (11).