RETRO-PROPAGATION APPROACH
| Collection | International Encyclopedia of Systems and Cybernetics |
|---|---|
| Year | 2004 |
| Vol. (num.) | 2(2) |
| ID | ◀ 2872 ▶ |
| Object type | Methodology or model |
The calculation of the quickest and most efficient path backward, from the goal to the actual configuration (C. HARDY, 2001, p. 49)
This author writes: “…whenever the final state of a system, or the solution to a problem, is known in advance, but the intermediate steps remain unknown, it is appropriate to use a procedure in which current events are guided by future outcomes”(Ibid)
Future planified outcomes install in fact a feedforward.
Hardy adds: “Some neural nets , for example, operate by retro-propagation , comparing, at each step, the current state with the final one, looking for the shortest route from the desired final state to the current one”(Ibid)
The retro-propagation approach leds to impressive success in chess-playing by computer.
See also
Epigenesis, Parallel distributed processing