TRAFFIC MONITORING WITH WISARD AND PROBABILISTIC LOGIC NODES

An earlier report applied the WISARD pattern recognition machine to discrimination of objects from traffic footage; simulations of a multi-layer network of weightless nodes, however, were found to be much less successful, due at least in part to the impoverished training set available to the simulation. Here, the same network simulations are compared with a simulation of the WISARD machine, allowing both to have identical training sets. The weightless multi-layer networks perform more successfully, suggesting that a hardware version of the networks, with capacity to use the full video footage as training input, would perform at least as well as the true WISARD machine in this sort of task.