We present, how neural networks and fuzzy-systems can be combined to improve or create rules, which consist of linguistic expressions represented by fuzzy-sets. Due to the fuzzy-component we are able to integrate and extract expert knowledge. The neural component is used for optimization with historical data by transforming the rule base into a special neural network architecture. The parameters of these neural network are optimized with gradient descent techniques, which are combined with a semantic preserving algorithm. Therefore the optimized parameters can be transformed into an improved and still interpretable rule base. The special architectures enables us to change the form of the fuzzy-sets and the structure of the rule base. For structural optimization we use so called priming techniques on premises and rules. Rules can be deleted or changed by deletion and/or insertion of single premises. Also the creation of semantically correct rules is possible using this techniques.
[1]
Detlef Nauck,et al.
Foundations Of Neuro-Fuzzy Systems
,
1997
.
[2]
Yann LeCun,et al.
Optimal Brain Damage
,
1989,
NIPS.
[3]
Volker Tresp,et al.
Network Structuring and Training Using Rule-Based Knowledge
,
1992,
NIPS.
[4]
Philip E. Gill,et al.
Practical optimization
,
1981
.
[5]
Ferdinand Hergert,et al.
Improving model selection by nonconvergent methods
,
1993,
Neural Networks.
[6]
Volker Tresp,et al.
Early Brain Damage
,
1996,
NIPS.