Superior Generalization Capabilities of Neuron-MOS Neural Networks in Mirror-Symmetry Problem Learning

We have studied the self-learning perforrnance of Neuron-MOS (UMOS) neural networks in solving mirror symmetry problems using computer simulation. Despite the inherent restrictions imposed on Hardware-Backpropagation (HBP) lcarning algorithm directly implernented on uMOS neural networks, a superior generalization capability (the ability to solve new problems not shown during the learning phase) has been demonstrated for HBP by optimwingthe circuit parameters. PC-4-6