The strength of these connections are known as the network's weights, and it is common to think of the network's "knowledge" as being stored in its set of weights. The values of these weights are modifiable, so, given some initial configuration, changes to the weights can be made that improve the performance of the network over time.
... the specific structure of the network, and the weight-adjustment algorithm, the network may learn to carry out some desired input-output mapping.
... most connectionist networks also exploit a distinctive kind of representation, so-called distributed representation, according to which a representation is conceived as a pattern of activation spread out across a group of processing units. p.10
... the specific structure of the network, and the weight-adjustment algorithm, the network may learn to carry out some desired input-output mapping.
... most connectionist networks also exploit a distinctive kind of representation, so-called distributed representation, according to which a representation is conceived as a pattern of activation spread out across a group of processing units. p.10
From this perspective, relevancy of information relates to its ability to change the network's weights. Irrelevant information passes by through the network without reconfiguring its "knowledge" and/or ability to act upon it.
Also of interest, a technique to build intelligence into a distribution sub-system. In this case, the distribution and control components of the system are integrated and cannot be substituted with a competing solution. The "vertical" axis (Distribution--Control) on the 5-element system diagram becomes not just important, but essential to the system's performance.
tags: cognition, network, control, computers, information, distribution, system, five element analysis
No comments:
Post a Comment