A non-classical approach to neural networks

Artificial neural networks have shown their usefulness in many applications. But they are hampered by some drawbacks such as long training times and limitations concerning implementation on low cost microcontrollers. As a possible solution, a non-classical network approach is presented, which is centered between symbolic and subsymbolic computation. It consists of nodes based on a lookup table. The node and network structures are discussed in more detail while a heuristic learning scheme is only outlined. Some examination results are presented using a closed loop control application