Hybrid Intelligent Systems
Hybrid systems or hybrid intelligent systems [Kandel92, Medsker94, Nikolopoulos97, Sun97, Cloete00] are computational systems which are based mainly on the integration of soft-computing techniques (especially artificial neural networks and fuzzy systems) but which also allow a “traditional” symbolic interpretation or interaction with symbolic components (like expert systems or knowledge-based systems).
These hybrid intelligent systems attempt to integrate different soft-computing techniques (known as computational intelligence tools) [Ebergart90, Furuhashi01] like fuzzy logic, neurocomputing, evolutionary computing, probabilistic computing, chaotic computing, and machine learning, exploring their complementarities in order to achieve better results for a given task. The rapidly growing complexity and sophistication of modern information/intelligent systems makes it clear that the achievement of success requires full exploration of the capabilities of each technique in our possession.
There are many possible combinations among the symbolic systems and soft-computing techniques, and also different ways to integrate them. For example, neural networks (NN) can be combined with fuzzy logic (FL), case-based reasoning (CBR), or genetic algorithms (GA) in order to obtain a unified model or even a co-processing scheme.
Hybrid Intelligent Systems
The main argument, which is the most used one, to justify the application of hybrid intelligent systems is based on the idea that most artificial intelligence (AI) techniques are complementary. Hybrid systems take advantage of their respective component strengths in order to increase the overall system performance and to eliminate the drawbacks of the components.
The representation
and manipulation of large amounts of knowledge in computers ensuring their integrity,
consistency, and effective exploitation is one of the main issues in AI research.
The knowledge representation format we choose should be able to handle the naturally
inaccurate and incomplete characteristics of data. Different knowledge representation
approaches (e.g., NN, FL, probabilistic computing, logical systems) have particular
ways of handling uncertain,
imprecise, or incorrect data. Some methods are more suitable to handle numerical
data, while others, symbolic data. There is no perfect knowledge representation
method that is able to represent all types of knowledge: numerical and discrete,
symbolic and non-symbolic, exact and inexact, precise and uncertain, specific
and general, etc.
| Knowledge - different types of representation and reasoning methods | |
| Natural Language – a powerful knowledge representation tool | |
| Question: What time will the movie “A.I.” start showing? | |
| Possible Answers: | |
| 1. The movie starts at 8:15 pm. | [Exact] |
| 2. The movie starts between 8:00 and 9:00 pm. | [Imprecise] |
| 3. I think the movie starts at 8:00 pm but I’m not really sure. | [Uncertain] |
| 4. The movie starts at around 8:00 pm. | [Approximate] |
| 5. It is possible that this movie starts at 8:00 pm. | [Possible] |
| 6. Probably the movie starts at 8:00 pm. [90% of probability] | [Probable] |
| 7. John said this movie starts at 8:00 pm. Mary said this movie starts at 10:00 pm. | [Inconsistent] |
| 8. I don’t know exactly at which hour the movie starts, but usually, it opens at 9:00 pm. | [Incomplete] |
| 9. I really don’t know. | [Unknown] |
| 10. I’m sure that this movie doesn’t start in the morning. | [Negation] |
| 11. The movie starts two hours after the previous session. | [Relative] |
| 12. The movie starts today at 8:00 pm and tomorrow at 9:00 pm | [Alternative] |
| 13. The movie sessions are at 6:00, 8:00, and 10:00 pm. This movie can only start at one of those hours. | [Options]
(Non-numerical) |
| 14. The movie starts only when the director arrives. | [Association] |
| 15. Next month. | [Temporal]
(Different scale) |
| … | ... |
Nevertheless, human knowledge is often difficult to express in a computer format. Typically, a specialist is unable to formulate perfectly her/his knowledge and experience in a specific knowledge representation language because human experts do not usually apply a unique knowledge representation and manipulation method. Humans solve problems using several “reasoning tools” such as inference, analogy, and deduction based on their own experience or on some previously acquired knowledge. Thus, natural human intelligence is based on multiple and hybrid knowledge representation and manipulation schemes.
Hybrid intelligent systems, for example neuro-symbolic systems, are able to integrate different representation and reasoning techniques in the same system. These systems mostly allow us to use information obtained from several different sources. Neuro-symbolic systems can also handle numerical and symbolic information more effectively than the individual systems do.
Briefly, hybrid systems target to improve AI tools in order to:
