Abstract
A generic architecture for realizing various neural network models is proposed. The connectivity problem is avoided by shifting the inputs to the neurons in a semiparallel manner using charge-coupled-device arrays. The synaptic interaction matrix can be loaded into the device in parallel. The proposed device can be implemented with present-day technologies.
| Original language | English |
|---|---|
| Pages | iii/403-409 |
| State | Published - 1987 |
| Externally published | Yes |