The authors present a digital neural model that permits high-density implementation of an integrated circuit with completely programmable weights. The digital model is derived from general analog neural models by analyzing their logical characteristics. After digitizing the analog neuron, the weights are expressed using logical values without losing the capabilities of the analog neurons. This permits the weights to be used as inputs; consequently, the logical values of the weights can be dynamically changed and also can be implemented with common discrete logics. To utilize the capability of dynamically variable weights, a learning algorithm was implemented in a network of digital neurons. The learning algorithm and simulation studies are briefly discussed.
|Original language||English (US)|
|Number of pages||6|
|Journal||Conference Proceedings - IEEE SOUTHEASTCON|
|State||Published - Apr 1 1989|
|Event||Energy and Information Technology in the Southeast - Columbia, SC, USA|
Duration: Apr 9 1989 → Apr 12 1989