System for self-organization of stable category recognition codes for analog input patterns

• US 5,133,021 A
• Filed: 02/28/1990
• Issued: 07/21/1992
• Est. Priority Date: 06/19/1987
• Status: Expired due to Term

• Current Assignee: Boston University
• Original Assignee: Boston University
• Inventors: Carpenter, Gail, Grossberg, Stephen
• Primary Examiner(s): Boudreau, Leo H.
• Application Number:
  US07/486,095
• Time in Patent Office: 874 Days
• Field of Search: 382/14, 382/15, 364/513, 364/274.9
• US Class Current: 382/156
• CPC Class Codes: G06F 18/23211 with adaptive number of clusters
• CPC Class Codes: G06N 3/0409 Adaptive resonance theory [ART] networks

Abstract

A neural network includes a feature representation field which receives input patterns. Signals from the feature representative field select a category from a category representation field through a first adaptive filter. Based on the selected category, a template pattern is applied to the feature representation field, and a match between the template and the input is determined. If the angle between the template vector and a vector within the representation field is too great, the selected category is reset. Otherwise the category selection and template pattern are adapted to the input pattern as well as the previously stored template. A complex representation field includes signals normalized relative to signals across the field and feedback for pattern contrast enhancement.

Claims

1. A pattern recognition system comprising:

a short term feature representation field of nodes storing signals of a feature pattern derived from input signals defining an input pattern, the short term input signals defining an input including means responsive to signals across the field for enhancing and repressing individual ones of said signals;

a category representation field of nodes;

long term memory storing weights;

means for selection at lest one category node of the category representation field based on a feature pattern from the short term feature representation field weighted by weights derived from long term memory; and

means for resetting a category selection such that another category node is selected based on the feature pattern.

2. A pattern recognition system as claimed in claim 1 further comprising means for adapting weights stored in long term memory to the input pattern.

3. A pattern recognition system as claimed in claim 1 further comprising means for generating a long term memory template, based on a selected category, to be matched against the input pattern.

4. A pattern recognition system as claimed in claim 3 further comprising means for adapting the long term memory template to the input signals.

5. A pattern recognition system comprising:

a short term feature representation field of nodes storing signals of a feature pattern derived from input signals defining an input pattern and template signals defining a long term memory template, the short term feature representation field including means responsive to signals across the field for enhancing and repressing individual ones of said signals;

a category representation field of nodes;

means for selecting at least one category of the category representation field based on a feature pattern from the short term feature representation field;

means for generating the long term template based on the at least one selected category; and

means for adapting category selection and the long term memory template to the input signals.

6. A pattern recognition system as claimed in claim 5 wherein at least one category is selected based on a pattern from the short term feature representation field weighted by weights derived from long term memory, the system further comprising means for resetting a category selection such that another category node is selected based on the feature pattern.

7. A pattern recognition system comprising:

a short term feature representation field of networks of computational nodes storing signals of a feature pattern derived from input signals defining an input pattern, each network storing short term signals based on and specific to an input signal;

a category representation field of nodes;

long term memory storing weights;

means for selecting at least one category node of the category representation field based on a feature pattern from the short term feature representation field weighted by weights derived from long term memory; and

means for resetting a category selection such that another category node is selected based on the feature pattern.

8. A pattern recognition system as claimed in claim 7 further comprising means for adapting weights set by long term memory to the input pattern.

9. A pattern recognition system as claimed in claim 7 further comprising means for generating a long term memory template, based on a selected category, to be matched against the input pattern.

10. A pattern recognition system as claimed in claim 9 further comprising means for adapting the long term memory template to the input signals.

11. A pattern recognition system comprising:

a short term feature representation field of networks of computational nodes storing signals derived from input signals defining an input pattern and template signals defining a long term memory template, each network storing short term signals based on and specific to an input signal;

a category representation field of nodes;

means for selecting at least one category node of a category representation field based on a feature pattern from the short term feature representation field;

means for generating the long term template based on the at least one selected category; and

means for adapting category selection and the long term memory template to the input signals.

12. A pattern recognition system as claimed in claim 11 wherein at least one category is selected based on a pattern from the short term feature representation field weighted by weights derived from long term memory, the system further comprising means for resetting a category selection such that another category node is selected based on the feature pattern.

13. In a pattern recognition device comprising:

a short term feature representation field of nodes storing signals derived from input signals defining an input pattern and template signals defining a long term memory template,

a category representation field of nodes;

means for selecting at least one category node of the category representation field based on a feature pattern from the short term feature representation field,

means for generating the long term template based on a selected category,

means for adapting category selection and the long term memory template to the input signals, and

means for resetting a category selection, such that another category is selected based on the feature pattern, with an insufficient match between the input pattern and template,

the improvement wherein the short term feature representation field comprises networks of computational nodes, each network storing short term signals based on and specific to an input signal (I_i).

14. A method of pattern recognition comprising:

storing signals of a feature pattern in a short term feature representation field of nodes, the signals being derived from input signals defining an input pattern and individual ones of such signals being enhanced and repressed responsive to signals across the field;

storing weights in long term memory;

selecting at least one category node of a category representation field based on a feature pattern from the short term feature representation field weighted by weights derived from long term memory; and

resetting a category selection such that another category node is selected based on the feature pattern.

15. A method as claimed in claim 14 further comprising adapting weights stored in long term memory to the input pattern.

16. A method as claimed in claim 14 further comprising generating a long term memory template, based on a selected category, and matching the long term memory template against an input pattern.

17. A method as claimed in claim 16 further comprising adapting a long term memory template to the input signals.

18. A method of pattern recognition comprising:

storing feature signals of a feature pattern derived from input signals defining an input pattern and template signals defining a long term memory template and enhancing and repressing individual ones of said feature signals responsive to signals across a short term feature representation field;

selecting at least one category of a category representation field based on a feature pattern from the short term feature representation field;

generating the long term template based on the at least one selected category; and

adapting category selection and the long term memory template to the input signals.

19. A method of pattern recognition as claimed in claim 18 further comprising selecting the at least one category based on a pattern from the short term feature representation field weighted by weights derived from long term memory and resetting a category selection such that another category node is selected based on the feature pattern.

20. A method of pattern recognition comprising storing feature signals of a feature pattern derived from input signals defining an input pattern in a short term representation field of networks of computational nodes, each network storing short term signals based on and specific to an input signal;

storing weights in long term memory;

selecting at least one category node of a category representation field based on a feature pattern from the short term feature representation field weighted by weights derived from long term memory; and

resetting a category selection such that another category node is selected based on the feature pattern.

21. A method as claimed in claim 20 further comprising adapting weights set by long term memory to the input pattern.

22. A method as claimed in claim 20 further comprising generating a long term memory template, based on a selected category and matching the long term memory template against the input pattern.

23. A method as claimed in claim 22 further comprising adapting the long term memory template to the input signals.

24. A method of pattern recognition comprising:

storing signals derived from input signals defining an input pattern and template signals defining a long term memory template in a short term feature representation field of networks of computational nodes, each network storing short term signals based on and specific to an input signal;

selecting at least one category node of a category representation field based on a feature pattern from the short term feature representation field;

generating the long term template based on the at least one select category; and

adapting category selection and the long term memory template to the input signals.

25. A method as claimed in claim 24 further comprising selecting the at least one category based on a pattern from the short term feature representation field weighted by weights derived from long term memory and resetting a category selection such that another category node is selected based on the feature pattern.

26. A method of pattern recognition comprising:

storing signals derived from input signals defining an input pattern and template signals defining a long term memory template in a short term feature representation field of nodes;

selecting at least one category node of a category representation field based on a feature pattern from the short term representation field;

generating the long term template based on a selected category;

adapting category selection and the long term memory template to the input signals; and

resetting a category selection such that another category is selected based on the feature pattern with an insufficient match between the input pattern and long term memory template;

the improvement comprising storing signals of the feature representation field in networks of computational nodes, each network storing short term signals based on and specific to an input signal.