A new neural network approach to spatiotemporal pattern recognition

Concerns dynamic pattern recognition by neural net. There are several research currents: 1. Static network use together with effective technical methods of preprocessing signals; 2. Basic architecture modification in order to cause the network state dependence on prehistory; 3. Setting up of dynamic networks with internal capacity of context information comprehension and the use of the well-known neuroparadigms only as net elements of complex architecture. The approach of our research is based on the biological prototype's principles. They are: three-level functioning; series-parallel analysis and synthesis of ascending and descending information; afferent and efferent matching. A receptive field environment and its structure are defined, and a theoretical aspect description of dynamic neural network set-up for spatiotemporal pattern processing and computer simulation, ABCnet, is given.