A locally-connected multilayer neural network model with enzymatic neurons as basic processing elements is proposed. There exists internal dynamics of the Hopfield circuit in the enzymatic neuron which can be described by differential equations and the firing rule. There are two different types of connection weights. The connection weights related to the internal dynamics can be trained by using the Hebbian rule, and that related to the enzymatic configurations can be trained through evolutionary learning. This model can be used for pattern classification or associative memory. In the simulation study of pattern classification, the authors discover that the internal dynamics plays an important role in improving the noise-tolerance.
展开▼
