LEARNING MACHINE
| Collection | International Encyclopedia of Systems and Cybernetics |
|---|---|
| Year | 2004 |
| Vol. (num.) | 2(1) |
| ID | ◀ 1887 ▶ |
| Object type | Methodology or model |
- “A machine capable of performing certain actions even though the builders do not know algorithms for these actions” (P. DENNING, 1992a, p.l4).
The learning machine constructs its own algorithm to perform the task.
DENNING explains that it does so by adjusting “internal parameters of an associative memory”, as it registers successive examples “each consisting of an input and a corresponding ideal” (Ibid.).
He adds: “…the machine's internal structure can be represented by a set of rules telling it how to respond to given inputs”.
Moreover: “After each action, the machine uses the resulting pay-off feedback to modify its rule set so that an effective behavior is reinforced or an ineffective behavior is dropped”.
In this way “… a population of rules evolves over time and the rules producing the highest pay-offs come to dominate the population”. Thus “Genetic algorithms are being used as the builders of programs inside learning machines” (Ibid).
In order to be able to perform a task, a learning machine needs thus an algorithm producing algorithms, i.e. a genetic algorithm. Such a meta-algorithm must by necessity be able to produce, experiment and select combinations of basic rules.
The concept apply also quite probably to natural learning machines.