| Autor: |
Araki, H., Brézin, E., Ehlers, J., Frisch, U., Hepp, K., Jaffe, R. L., Kippenhahn, R., Weidenmüller, H. A., Wess, J., Zittartz, J., Beiglböck, W., Lehr, Sabine, Parisi, Jürgen, Müller, Stefan C., Zimmermann, Walter, Ruwisch, D., Bode, M., Schulze, H.-J., Niedernostheide, F.-J. |
| Zdroj: |
Nonlinear Physics of Complex Systems; 1997, p194-212, 19p |
| Abstrakt: |
We present synergetic hardware concepts for self-organizing feature representations, i.e., for the Kohonen map, the "neural-gas" vector quantization network, and generalizations of these basic procedures. Ignition of fronts and filaments in suitably designed semiconductor devices are universal synergetic mechanisms proposed to implement the interactive parts of the algorithms. Basically, the best matching unit ("winner neuron") is determined in parallel by the (first) ignition. Learning is controlled by the intrinsic dynamics of the semiconductor device, i.e., either by front propagation or by successive ignition of further filaments when a global control parameter such as the supply voltage is increased. The architectures presented in this article operate in a highly parallel manner whereas connectivity within the network is maintained very low. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Supplemental Index |
| Externí odkaz: |
|
