Information and Knowledge Representation in Fusion Problems - Information Fusion in Signal and Image Processing - page 56

- they do not model exactly the same concepts or the same aspects of the informa-
tion;
- they do not have the same semantics;
- they do not have the same power of representation;
- they do not have the same reasoning power.
In particular, the first two points make it illusory and misleading to want to com-
pare their performances on the same applications 1 .
These remarks are also a motivation for hybrid representation techniques, which
allow us to simultaneously represent elements of information with different types of
imperfections. It is also possible to define the probabilities of fuzzy events, of the
belief functions of fuzzy subsets, etc. However, these approaches are still rarely used
in information fusion.
5.4. Symbolic representation of imperfect knowledge
Artificial intelligence is traditionally defined (in Minsky's and McCarthy's works,
for example) from two points of view:
- from the cognitive point of view: this consists of constructing computable mod-
els of cognitive processes, in other words programs that can simulate human perfor-
mances;
- from the computer science and engineering point of view: this consists of assign-
ing to computers tasks that would be considered intelligent if performed by a human,
in other words extending the abilities of computers.
Artificial intelligence generated symbolic representations of knowledge. The field
of knowledge representation is characterized by:
- the definition of a representation as a set of syntactic and semantic rules to
describe an element of knowledge;
- logical representations (the expressivity depends on the logic used; see section
5.6);
- compact representations (only the relevant and characteristic properties are
stated explicitly);
- ease of use;
- what is important is actually explicit.
1. This partly explains the contradictory conclusions found on this subject in other works.