Information Technology Reference
In-Depth Information
Fig. 6 Example network with two higher-level concepts A-B-C-D and C-D-E (Greer
2008
)
same idea as the indexing system being used here and a diagram of this is shown in
Fig.
6
.
The nodes are meant to represent concepts and groups of them, higher-level
concepts. However, because there can be overlap between concepts they can be
grouped together, with different indexes de
ning each exact group. If concept trees
were used, a tree consisting of A-B-C-D could link to a tree consisting of E only,
for example. The ABCD tree would have a base node with some value and then
branches, one of
. An event entity would then need to
activate the base node of the tree and activate all of its branches, to realise the
'
AtoB
'
and one of
'
CtoD
'
rst
concept group. To realise the second group, a different event entity would need to
link to and activate both the ABCD tree and the E tree, at the same time. Then
possibly and interestingly, can a link between the two trees themselves complete a
circuit, to indicate the other concept group of CDE. If a link between the leaf D
node and the base E tree node exists, for example? This might be a more dynamic
model than the original design of Fig.
6
that considered
fixed sets of (unique)
reinforced links only. The key sets possibly sit on-top of the linking structure,
where both can change dynamically. So there are two different possibilities for
dynamic change, but with the new functionality, there are also other technical
dif
culties. The intention is that concept groups will represent something more
meaningful and therefore can be used as part of a reasoning process. This paper
would suggest that it is more of a memory structure, but with the same goal of
de
ning higher-level concepts more accurately, to allow them to be reasoned over.
7.3 Complementary Structures
Dynamic links have therefore been used previously (Greer
2008
,
2011
) as part of a
neural network, but the two techniques are probably compatible. It is curious that
the knowledge-based concept tree, in relation to the 3-level cognitive model
described in Greer (
2008
, Sect. 9.3.8 or
2013b
), would be more closely associated
with the
first optimising level and the second aggregating level. It would create the
base or bed of the system. The experience-based neural network would then be
more closely associated with the third level. It would manipulate the knowledge
(cleverly) through a dynamic, experience-based approach. Looking at the concept
trees has actually helped to create a clearer picture and provide some more con-
sistency over the whole model. If the concept trees are used to create pattern groups
Search WWH ::
Custom Search