Nonoverlapped trees of probabilistic logic neurons

The working principles of nonoverlapped trees of probabilistic logic neurons (NOTPLN) are discussed. Three learning algorithms for NOTPLN are described. Simulation experiments are used to demonstrate that NOTPLN is insensitive to noise and is able to generalize the rules behind the training set. In this simulation, NOTPLN is trained with a set of road scenes as the input pattern and the corresponding actions for the driver as the output pattern. A conflict reduction algorithm is used. The road scene is modeled as one dimension. It is a horizontal line cut across the road and the pavement at both sides. The set of training patterns is shown.<
>