Matches in SemOpenAlex for { <https://semopenalex.org/work/W2241551256> ?p ?o ?g. }
Showing items 1 to 65 of
65
with 100 items per page.
- W2241551256 abstract "The primary aim of this research is to investigate a general optimal algorithm of rule extraction from a neural network and then to apply this rule extraction method to the neural tree net and the SIR (simultaneous induction of rules) net. The neural tree net is a multifeature split, decision tree based network. The SIR is a new neural network based learning methodology. This rule extraction process translates knowledge from an internal representation of a neural network system to symbolic production rules. The proposed rule extraction method yields production rules or concept descriptions that are syntactically similar to the rules generated by the symbolic learning approach. This allows for a better understanding of neural network learning and an easier integration of neural network technology with the expert systems. The individual rules of the proposed methodology are formed on the basis of high-level attributes which are automatically discovered during the neural learning process.The neural tree net builds the decision tree by finding the best hyperplanes in a top-down manner and by using either the modified pocket algorithm or the AMIG (average mutual information gain) delta rule. The SIR net builds the decision tree by finding the hyperplanes in a parallel manner by using the backpropagation algorithm and the winner-takes-all strategy. Both systems incorporate the best features of the symbolic and neural learning approaches. Furthermore, in the SIR system, the use of neural learning permits the simultaneous induction of rules because of the parallelism inherent in the neural networks. This provides for a faster degree of learning in the SIR net. The simultaneous induction of rules also avoids the problem of under or over learning that is typically present in the sequential, top-down inductive methods.The proposed rule extraction method can apply to any feedforward neural networks designed for symbolic rule extraction. Thus, this research provides an important step in removing the opaqueness of neural networks which has bottlenecked the neural network research in the past." @default.
- W2241551256 created "2016-06-24" @default.
- W2241551256 creator A5083832200 @default.
- W2241551256 date "1993-01-11" @default.
- W2241551256 modified "2023-09-23" @default.
- W2241551256 title "Symbolic rule extraction from artificial neural networks" @default.
- W2241551256 hasPublicationYear "1993" @default.
- W2241551256 type Work @default.
- W2241551256 sameAs 2241551256 @default.
- W2241551256 citedByCount "3" @default.
- W2241551256 crossrefType "journal-article" @default.
- W2241551256 hasAuthorship W2241551256A5083832200 @default.
- W2241551256 hasConcept C10229987 @default.
- W2241551256 hasConcept C113174947 @default.
- W2241551256 hasConcept C119857082 @default.
- W2241551256 hasConcept C134306372 @default.
- W2241551256 hasConcept C154945302 @default.
- W2241551256 hasConcept C155032097 @default.
- W2241551256 hasConcept C175202392 @default.
- W2241551256 hasConcept C177973122 @default.
- W2241551256 hasConcept C33923547 @default.
- W2241551256 hasConcept C41008148 @default.
- W2241551256 hasConcept C50644808 @default.
- W2241551256 hasConcept C5481197 @default.
- W2241551256 hasConcept C84525736 @default.
- W2241551256 hasConceptScore W2241551256C10229987 @default.
- W2241551256 hasConceptScore W2241551256C113174947 @default.
- W2241551256 hasConceptScore W2241551256C119857082 @default.
- W2241551256 hasConceptScore W2241551256C134306372 @default.
- W2241551256 hasConceptScore W2241551256C154945302 @default.
- W2241551256 hasConceptScore W2241551256C155032097 @default.
- W2241551256 hasConceptScore W2241551256C175202392 @default.
- W2241551256 hasConceptScore W2241551256C177973122 @default.
- W2241551256 hasConceptScore W2241551256C33923547 @default.
- W2241551256 hasConceptScore W2241551256C41008148 @default.
- W2241551256 hasConceptScore W2241551256C50644808 @default.
- W2241551256 hasConceptScore W2241551256C5481197 @default.
- W2241551256 hasConceptScore W2241551256C84525736 @default.
- W2241551256 hasLocation W22415512561 @default.
- W2241551256 hasOpenAccess W2241551256 @default.
- W2241551256 hasPrimaryLocation W22415512561 @default.
- W2241551256 hasRelatedWork W131552660 @default.
- W2241551256 hasRelatedWork W1518494427 @default.
- W2241551256 hasRelatedWork W1599216479 @default.
- W2241551256 hasRelatedWork W1601698119 @default.
- W2241551256 hasRelatedWork W1650405992 @default.
- W2241551256 hasRelatedWork W181107822 @default.
- W2241551256 hasRelatedWork W1989574962 @default.
- W2241551256 hasRelatedWork W2036677213 @default.
- W2241551256 hasRelatedWork W2065204741 @default.
- W2241551256 hasRelatedWork W2072199831 @default.
- W2241551256 hasRelatedWork W2088203270 @default.
- W2241551256 hasRelatedWork W2108704760 @default.
- W2241551256 hasRelatedWork W2120299300 @default.
- W2241551256 hasRelatedWork W2235915642 @default.
- W2241551256 hasRelatedWork W2349735111 @default.
- W2241551256 hasRelatedWork W2371789596 @default.
- W2241551256 hasRelatedWork W2413313105 @default.
- W2241551256 hasRelatedWork W2542042057 @default.
- W2241551256 hasRelatedWork W2899983622 @default.
- W2241551256 hasRelatedWork W3147413282 @default.
- W2241551256 isParatext "false" @default.
- W2241551256 isRetracted "false" @default.
- W2241551256 magId "2241551256" @default.
- W2241551256 workType "article" @default.