Human-in-the-loop control with majority vote neural networks

The specifications for many automated decision-making/aiding systems include a human censor in the loop. However, the problem of human cognitive overload that arises in highly complex situations necessitates that the human be relieved of much of the lower level data. The authors present a simple, robust neural network for self-organized learning that hierarchically recognizes successively higher objects from patterns in the input sensor signals. It is called the majority vote neural network. These objects are decoded into a situation-response frame for presentation to the human. If the human approves the frame, it goes to the command sequence generator to be decoded further into a sequence of commands to drive the required actions. Otherwise, the human must supply an alternate response codeword to the situation-response frame