Generally, a expert system is a tool able to reproduce the cognitive mechanisms of an expert, in a particular field. It is about the one of the ways trying to lead to the Artificial intelligence.
More precisely, an expert system is a Logiciel able to answer questions, by carrying out a reasoning starting from known facts and rules. It can be used in particular like tool as Decision-making aid. The first expert system is DENDRAL. It made it possible to identify the chemical components.
An expert system is composed of 3 parts:
The inference engine is able to use made and rules to produce new facts, until arriving at the answer to the put expert question.
The majority of the existing expert systems rest on mechanisms of formal Logique (logic aristotelician) and use the deductive reasoning. Essentially, they use the rule of following inference (Syllogisme):
- if P is true ( made or premise ) and if it is known that P implies Q (rule) then , Q is true ( new fact or conclusion).
Simplest of the expert systems rest on the Logique proposals (known as also “ logical of order 0 ”). In this logic, one uses only proposals, which are true, or false. Other systems rest on the Logique predicates first order (known as also “ logical of order 1 ”), that algorithms make it possible to handle easily.
Lastly, to facilitate the description of real problems in the form of logical rules, one has recourse to additional operators or values (concepts of need/possibility, coefficients of plausibility, etc).
Inference engine
There exist many driving types of , able to treat various forms of logical rules to deduce from new facts starting from the Base of knowledge.One often distinguishes three categories, based on the way in which the problems are solved:
- the driving - known as with “ chaining before ” - which leaves the facts and rules the Base knowledge, and try to approach the facts sought by the problem.
- the engines - said to “ chaining postpones ” - which leaves the facts sought by the problem, and try via the rules, “to go back” to known facts,
- the engines - said to “ mixed chaining ” - which use a combination of these two approaches chaining before and chaining postpones .
Some driving d" inferences can be partially controlled or controlled by méta-rules which modify their operation and their methods of Raisonnement.
Acquisition of knowledge
If the algorithms of handling of facts and rules many and are known, the determination of the whole of the facts and rules which will compose the base of knowledge is delicate problems. How to describe the behavior of an expert vis-a-vis a particular problem, and its manner of solving it, there is the question. Because what one wishes to obtain is neither more neither less than the experiment, the practical knowledge of the expert, and not the theory which one can exclusively find in the books nor the logical rules of inference. Equivalents of the methods of analysis of traditional data processing, of the methods of acquisition of knowledge are developed.
History
The first expert systems are born in the USA in the years 1980. MYCIN , which handled expertise in the medical field, is one of most known. They had their hour of glory in the years 1980, where it was too quickly thought that they could develop massively. In practice, the development of this kind of application is very heavy because, starting from a hundred simple rules, there is enormously evil to include/understand how the expert system “reasons” (handles made and rules in real-time), and thus to ensure the final settling of it then maintenance. The project SACHEM (piloting of blast furnace at Arcelor), operational in the years 1990, is one of the last projects “expert system” resulting from research to have been born. Today, of multiples small expert systems are operational in industry and the services without one speaking about it. One prefers to acknowledge to use knowledge bases as that which Microsoft puts on line for its products.
See too
External bond
- prolog
- Expert system CLIPS
- Expert system JESS (Java Expert System Shell)
- ProBT® inference engine probabilistic in commercial release at ProBAYES and free for research and teaching on the site Bayesian-Programming.org
Simple: Expert system
| Random links: | Pokédex | Subway of Madrid | Valle Mosso | Purple Webb | Remi-Pierre Paquin |