An aspect that appears to be of great importance in human decision making is focus of attention. This focus determines the level of detail that should be considered in addressing the current situation. Classification neural networks as they currently exist generally rely on building an overall model based on the data presented. Implementation of a level of detail structure depends on hierarchical modeling. Neural networks at each level of detail must be trained separately, with each requiring different data sets for training and testing. In addition, a method for deciding which level is appropriate must be developed. In the work described in this paper, meta knowledge, a technique derived from knowledge-based reasoning, is used for transition between multiple levels. The meta knowledge described internally structures transitions among the neural network layers.
[1]
Donna L. Hudson,et al.
Comparison of the hypernet learning algorithm with other neural network approaches
,
1998,
Computers and Their Applications.
[2]
Donna L. Hudson,et al.
Approaches to management of uncertainty in an expert system
,
1988,
Int. J. Intell. Syst..
[3]
M. Cohen,et al.
Comparative Approaches to Medical Reasoning
,
1995
.
[4]
Ronald R. Yager,et al.
Approximate reasoning as a basis for rule-based expert systems
,
1984,
IEEE Transactions on Systems, Man, and Cybernetics.
[5]
Donna L. Hudson,et al.
A HYBRID NEURAL NETWORK WITH SYMBOLIC ACTION LAYER FOR MEDICAL DECISION SUPPORT
,
1995
.