Improved recognition capabilities for goal seeking neuron

RAM based neural networks are a relatively new class of neural network which exhibit faster learning and greater ease of VLSI implementation than the traditional analogue models. Two RAM based neural models, the probabilistic logic neuron (PLN) and the goal seeking neuron (GSN), are simulated to determine their recognition capabilities. It is found that the PLN has very poor capabilities, whereas the GSN has widely varying capabilities due to the random nature of the GSN learning algorithm. A new GSN learning algorithm is presented which gives consistently good results.