Knowledge representation using geometric flexible adaptable templates a new machine learning methodology

• US 20030236761A1
• Filed: 04/06/2003
• Published: 12/25/2003
• Est. Priority Date: 04/08/2002
• Status: Active Grant

• Current Assignee: Tahoe Research Limited (f/k/a Learndale Limited) (Vector Capital Corporation)
• Original Assignee: Fiske Software LLC
• Inventors: Fiske, Michael Stephen
• Granted Patent: US 7,249,116 B2
• Time in Patent Office: Days
• US Class Current: 706/46
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

This invention relates to the machine recognition and learning of predetermined categories and more generally, to the representation of patterns, information and knowledge in computational applications. A method of learning categories is an important component of advanced software technology. This invention has applications in the following areas: bioinformatics, document classification, document similarity, financial data mining, goal-based planners, handwriting and character recognition, information retrieval, natural language processing, natural language understanding, pattern recognition, search engines, strategy based domains such as business, military and games, and vision recognition.

Claims

1. A machine comprising:

(a) a memory which stores a collection of templates,

(b) a learning component which constructs said templates,

whereby said machine does not get stuck in local minima, and whereby said machine enhances what any machine can learn, and whereby said machine greatly increases the speed, and whereby said machine enhances the representation of knowledge.

2. The machine of claim 1 wherein one or more category spaces are used.

3. The machine of claim 1 wherein one or more possibility spaces are used.

4. The machine of claim 1 wherein said templates are user-created.

5. The machine of claim 1 wherein evolution is used.

6. The machine of claim 1 wherein template maps are used.

7. The machine of claim 1 wherein matching functions are used.

8. The machine of claim 1 wherein template value functions are used.

9. The machine of claim 1 wherein prototype functions are used.

10. The machine of claim 1 wherein templates are removed from the template set.

11. The machine of claim 1 wherein shape recognition templates are used.

12. The machine of claim 1 wherein information retrieval templates are used.

13. The machine of claim 1 wherein bioinformatics templates are used.

14. A machine comprising:

(a) a memory which stores a collection of templates,

(b) a recognition component which utilizes said templates,

whereby said machine greatly increases the speed, and whereby said machine greatly increases the accuracy, and whereby said machine enhances the patterns, information and/or knowledge that any machine can retrieve, categorize, classify and/or recognize.

15. The machine in claim 14 wherein templates are chosen randomly.

16. The machine in claim 15 wherein a non-uniform probability distribution is used.

17. The machine in claim 14 wherein matching functions are used.

18. The machine in claim 14 wherein categories are removed during execution of said recognition component.

19. The machine in claim 14 wherein an example is in zero or more categories.

20. The machine in claim 14 wherein templates are user-created.

21. The machine in claim 14 wherein templates were constructed from evolution.

22. The machine of claim 14 wherein shape recognition templates are used.

23. The machine of claim 14 wherein information retrieval templates are used.

24. The machine of claim 14 wherein bioinformatics templates are used.

25. A machine comprising:

(a) evolution which builds competent elements from building blocks,

(b) a memory which stores said elements,

whereby said machine enhances the competency of said elements constructed, and whereby said machine greatly increases the speed at which said elements are constructed.

26. The machine in claim 25 wherein unbounded sequences are used.

27. The machine in claim 25 wherein sequences with different lengths are used.

28. The machine in claim 27 wherein sequence encoding at a location is variable.

29. The machine in claim 25 wherein said building blocks are functions.

30. The machine in claim 25 wherein said elements are selected from the group consisting of templates and components of templates.

31. The machine in claim 30 wherein the fitness of said elements depends on their ability to distinguish examples.

32. The machine in claim 25 wherein sequences encode a composition of building blocks.

33. The machine in claim 25 wherein the crossover locations are different.

34. The machine in claim 25 wherein the fitness of an element depends on its computational time.

35. The machine in claim 25 wherein the fitness of an element depends on its memory usage.

36. The machine in claim 25 wherein the fitness of an element depends on its competence.

37. A method of building a representation of patterns and/or information and/or knowledge, comprising:

(a) using a memory which stores a collection of templates,

(b) executing a learning phase which constructs said templates,

whereby said method does not get stuck in local minima, and whereby said method enhances what any method can learn, and whereby said method greatly increases the speed, and whereby said method enhances what can be represented.

38. The method of claim 37 wherein one or more category spaces are used.

39. The method of claim 37 wherein one or more possibility spaces are used.

40. The method of claim 37 wherein said templates are user-created.

41. The method of claim 37 wherein evolution is used.

42. The method of claim 37 wherein template maps are used.

43. The method of claim 37 wherein matching functions are used.

44. The method of claim 37 wherein template value functions are used.

45. The method of claim 37 wherein prototype functions are used.

46. The method of claim 37 wherein templates are removed from the template set.

47. The method of claim 37 wherein shape recognition templates are used.

48. The method of claim 37 wherein information retrieval templates are used.

49. The method of claim 37 wherein bioinformatics templates are used.

50. A method comprising:

(a) using a memory which stores a collection of templates,

(b) executing a recognition phase which utilizes said templates,

whereby said templates greatly increase the speed, and whereby said templates greatly increase the accuracy, and whereby said templates enhance the patterns, information and/or knowledge that any method can retrieve, categorize, classify and/or recognize.

51. The method in claim 50 wherein templates are chosen randomly.

52. The method in claim 51 wherein a non-uniform probability distribution is used.

53. The method in claim 50 wherein matching functions are used.

54. The method in claim 50 wherein categories are removed during execution of said recognition component.

55. The method in claim 50 wherein an example is in zero or more categories.

56. The method in claim 50 wherein user-created templates are used.

57. The method in claim 50 wherein templates were constructed from evolution.

58. The method of claim 50 wherein shape recognition templates are used.

59. The method of claim 50 wherein information retrieval templates are used.

60. The method of claim 50 wherein bioinformatics templates are used.

61. A means for information retrieval comprising:

(a) a memory which stores a collection of templates,

(b) a retrieval phase which retrieves information using said templates.

62. The means in 61 wherein one or more said templates use a <

DOER, ACTION>

phrase.

63. The means in 61 wherein one or more said templates use a <

DOER, ACTION, OBJECT>

phrase.

64. The means in 61 wherein one or more said templates are concept templates.

65. The means in 64 wherein said concept templates are selected from the group consisting of:

at least one of the following concepts occur;

all of the following concepts occur;

all of the following concepts occur in the same sentence;

all of the following concepts occur in this order;

all of the following concepts occur in the same sentence in this order;

none of the following concepts occur;

at least one of the following concepts do not occur;

all of the following concepts do not occur in the same sentence;

all of the following concepts do not occur in this order;

all of the following concepts do not occur in the same sentence in this order;

all of the following concepts occur in the same <

DOER, ACTION>

phrase;

all of the following concepts occur in the same <

DOER, ACTION, OBJECT>

phrase.