R Al-Alawi and T.J. Stonham
Synopsis
The digital neuron model is evaluated in terms of its functional capacity, generalisation, training procedure and hardware implementation, and contrasted with the analogue neuron model. A learning algorithm for digital multi-layer neural networks which uses a backpropagation search techniques is proposed and compared with other well-known training methods. Finally a dynamic mapping strategy for the nodes of the digital multi-layer network and the introduction of redundancy are evaluated as means of increasing the flexibility and functional capacity of the network.
Key words:
Boolean Neural Networks, Multi-Layer Networks, Perception, Probabilistic Logic Node, Learning Systems, Network Functional Capacity, Generalisation