Fuzzy neural networks

• US 6,192,351 B1
• Filed: 01/27/1998
• Issued: 02/20/2001
• Est. Priority Date: 02/24/1995
• Status: Expired due to Term

• Current Assignee: Genmark Diagnostics Incorporated (Roche Holding AG)
• Original Assignee: Osmetech PLC (Roche Holding AG)
• Inventors: Persaud, Krishna Chandra
• Primary Examiner(s): Hafiz, Tariq H.
• Assistant Examiner(s): DAVIS, GEORGE B
• Application Number:
  US08/913,472
• Time in Patent Office: 1,120 Days
• Field of Search: 706/31, 706/20, 706/16, 706/2, 382/156, 382/157, 382/159
• US Class Current: 706/2
• CPC Class Codes: G06F 2218/12 Classification; Matching
• CPC Class Codes: G06N 3/043 based on fuzzy logic, fuzzy membership or fuzzy inference, e.g. adaptive neuro-fuzzy inference systems [ANFIS]

Abstract

There is disclosed a pattern identifying neural network comprising at least an input and an output layer, the output layer having a plurality of principal nodes, each principal node trained to recognize a different class of patterns, and at least one fuzzy node trained to recognize all classes of patterns recognized by the principal nodes but with outputs set out at levels lower than the corresponding outputs of the principal nodes.

Claims

1. A pattern identifying neural network comprising at least an input and an output layer, the output layer having a plurality of principal nodes, each principal node trained to recognize a different class of patterns, and at least one fuzzy node trained to recognize all classes of patterns recognized by the principal nodes but with outputs set out at levels lower than the corresponding outputs of the principal nodes.

2. A neural network according to claim 1 which further comprises at least one hidden layer of nodes.

3. A neural network according to claim 2 in which a feedforward architecture is employed.

4. A neural network according to claim 3 in which the number of nodes in each hidden layer is equal to the number of nodes in the input layer plus a biasing node.

5. A neural network according to claim 3 or claim 4 in which the error back propagation algorithm is used to train the network.

6. A neural network according to claim 1 in which a fuzzy pattern classification system is employed.

7. A neural network according to claim 6 in which, providing the output from the fuzzy node is the largest nodal output in the output layer and does not exceed the output of at least one principal node by a predefined value, the principal node having the next greatest output is selected as representing the most likely class of patterns.

8. A neural network according to claim 7 in which the probability distribution representing the likelihood of an input pattern falling into any of the classes of patterns represented by each of the principal output nodes is calculated.

9. A neural network according to claim 1, in which the output layer comprises two slabs, each slab comprising a plurality of principal nodes and a fuzzy node, the principal nodes of one slab being trained to recognize global classes of patterns and the principal nodes of the other slab being trained to recognize sub-classes of patterns.

10. A neural network according to claim 9 in which the input pattern input to the input layer comprises the outputs of a plurality of gas sensors or quantities related thereto.

11. A neural network according to claim 10 in which the principal nodes of the second slab are trained to recognize patterns representing different concentrations of a least one gas or volatile species.

12. A neural network according to claim 10 in which the output of a temperature sensor is input to the input layer.

13. A neural network according to claim 10 in which the output of a humidity sensor is input to the input layer.

14. A neural network according to claim 1 in which the input pattern is reduced by a linear or non-linear mapping technique and the results therefrom, together with the unreduced pattern, are input to the input layer.

15. A neural network according to claim 14 in which the mapping technique is principal components analysis.

16. A neural network according to claim 1 in which the input pattern is preprocessed prior to pattern identification.

17. A neural network according to claim 16 in which the preprocessing comprises removal of elements of the input pattern which do not contribute substantially to the final identification.

18. A neural network according to claim 16 in which the preprocessing comprises decimation of the input database.