Image Processing Reference
In-Depth Information
In vision, specific tasks of focusing and adapting (with their attentional mecha-
nisms, for revision or repairs and consistency management), co-operation and fusion
(confrontational, augmentative, integrative) and co-ordination (deliberative, reactive,
optimal) are added to KBS. These methods are discussed in detail in [GAR 01] and
will not be covered here.
An example of a multi-agent system will be given in Chapter 10.
5.6. Reasoning modes and inference
There is a wider variety of inference modes used in KBS than in traditional knowl-
edge bases. These modes are divided into the following categories:
- deductive reasoning, which provides consequences based on facts (for example,
if the fact base contains
A
and the proposition
A
B
, then we can infer
B
);
- contraposition allows us to reason on non-observations (for example, if we have
→
A
B
and non-
B
, we can infer non-
A
);
- abductive reasoning attempts to find the causes explaining the observations (for
example, based on
A
→
→
B
and the observation of
B
, we infer that
A
is a possible
cause of
B
);
- inductive reasoning allows us to infer rules from regular or usual observations
(for example, if we have
B
every time we have
A
, we can infer
A
B
);
- projection provides consequences based on actions (if the fact base contains the
proposition
A
→
B
and we perform
A
, we expect
B
to occur);
- planning establishes which actions to perform in order to achieve goals (if we
want
B
and the base contains
A
→
→
B
, then we infer the action
A
).
These last two inference modes are particularly developed in embedded KBS, such
as those used in mobile robotics [SAF 02].
Information fusion often requires the help of different reasoning modes, in order
to better grasp and represent the nuances and subtleties of human reasoning.
In monotonic reasoning, obtaining more information naturally leads to more con-
clusions: if
A
is inferred from a base
KB
, we will also infer
A
from
KB
∪
B
. Tradi-
tional, propositional and first-order logic resort to this mode of reasoning.
In non-monotonic reasoning, new information can invalidate previous conclusions.
In the presence of imperfect knowledge and information, as is the case in information
fusion, sources of non-monotonicity essentially come from the hypotheses and restric-
tions that are applied. These are necessary to be able to reason, but can be questioned
if new information or elements of knowledge become available. These hypotheses that
are sources of non-monotonicity include:
Search WWH ::
Custom Search