NETWORK (Hopfield)
| Collection | International Encyclopedia of Systems and Cybernetics |
|---|---|
| Year | 2004 |
| Vol. (num.) | 2(2) |
| ID | ◀ 2266 ▶ |
| Object type | Methodology or model |
For a good introductory graphical description of this network, see P. DENNING (1992b, p.427-8).
HOPFIELD introduces nonlinear dynamics in computational networks. In DENNING's words: “HOPFIELD showed that his network can memorize patterns by storing them directly into the weights; no training is needed. There is a feedback path from the output of each unit to the input of every other unit, and all of these paths are assigned individual weights… Ultimately the network settles into a condition of equilibrium… If an input pattern does not exactly match one of the memorized patterns, the network will reach equilibrium at the nearest memorized pattern. This behavior provides a means for restoring a pattern from a noisy or a partial version” (Ibid).
HOPFIELD's networks are somehow reminiscent of ASHBY's homeostat.