Method and system for adaptive learning and pattern recognition

• US 20040205035A1
• Filed: 04/30/2001
• Published: 10/14/2004
• Est. Priority Date: 05/01/2000
• Status: Abandoned Application

• Current Assignee: Sensoria Technology Limited
• Original Assignee: Sensoria Technology Limited
• Inventors: Rimoux, Norbert
• Application Number:
  US09/845,381
• Publication Number:
  US 20040205035A1
• Time in Patent Office: Days
• US Class Current: 706/14
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06V 30/194 References adjustable by an adaptive method, e.g. learning
• CPC Class Codes: G06V 30/274 Syntactic or semantic context, e.g. balancing

Abstract

A method and system for an adaptive learning and pattern recognition that detects and learns significant patterns and pattern semantic in a noisy environment. A unique predation method of reinforcement and deletion by using stimulus and context activities is disclosed. The learning and pattern recognition of the present invention also employs the learned semantics knowledge to evaluate the semantic context of the pattern.

Claims

1. A method of recognizing and learning patterns in an adaptive learning network, comprising:

receiving an input pattern made up of one or more basic components, the one or more basic components measured in event time corresponding to an order of occurrence of the one or more basic components in the input pattern;

searching in an adaptive learning network for a cel structure having a basic component of the one or more basic components starting in the order of occurrence in the input pattern;

positively reinforcing the cel structure having the basic component;

creating one or more cel structures to hold the one or more basic components and linking the created one or more cel structures to the adaptive learning network according to its event time, if not found.

2. The method of claim 1, wherein the step of receiving includes buffering the input pattern.

3. The method of claim 1, wherein the method further includes negatively reinforcing one or more cel structures that have the same event time in the adaptive learning network as the positively reinforced cel structure.

4. The method of claim 3, wherein the method further include deleting the negatively reinforced one or more cel structures when the one or more cel structures have been negatively reinforced to a predetermined threshold.

5. The method of claim 1, wherein the step of creating includes creating one or more cel structures to hold the one or more basic components and linking the created one or more cel structures to a short term associated memory according to its event time, if not found.

6. The method of claim 5, wherein the method further includes migrating the short term associated memory into the adaptive learning network when the one or more cel structures in the short term associated memory have been positively reinforced to a predetermined threshold.

7. The method of claim 5, wherein the method further includes creating a semantic network linked to the one or more cel structures in the adaptive learning network, the semantic network including one or more terms that are semantically related to the linked one or more cel structures.

8. The method of claim 7, wherein the method further includes

searching the semantic network to further evaluate the input pattern in a semantic context.

9. The method of claim 5, wherein the method further includes migrating one or more cel structures in the short term associate memory to the adaptive learning network when the one or more cel structures in the short term associated memory have been positively reinforced to a predetermined threshold.

10. An adaptive learning and pattern recognition system, comprising:

a plurality of cel structures, each of the plurality of cel structures enabled to link to any one or combination of a parent cel structure, a cousin cel structure, and a child cel structure, wherein a parent cel structure contains an event in a pattern that occurred at event time t−

1, the cousin cel structure contains an event in a pattern that occurred at even time t, and the child cel structure contains an event in a pattern that occurred at event time t+1;

each of the plurality of cel structures enabled to positively reinforce itself when an input event matches its content, each of the plurality of cel structures enabled to create a cousin cel when an input event does not match its content,

wherein the plurality of cel structures are interrelated with one another in a chronological order of occurrence of events in a pattern forming one or more paths representing one or more patterns.

11. The system as claimed in claim 10, wherein each of the plurality of cel structures further has one of termination attribute and glue attribute;

and

a cel structure with termination attribute further has a link to a semantic network.

12. The system as claimed in claim 10, wherein each of the plurality of cel structures further includes:

a stimulus activity representing an activity of a cel in relation to the event in the signal that appears at the same event time as the cel in its path; and

a context activity representing a mean value of the stimulus activity and a context activity inherited from its parent cel in the same path.

13. The system as claimed in claim 12, wherein a degree of recognition is determined by a context activity computed for a cel structure with termination attribute in a path.

14. The system as claimed in claim 10, wherein the cel structure with termination attribute further includes a reference to a procedure for further identifying a pattern.

15. A program storage device readable by machine, tangibly embodying a program of instructions executable by the machine to perform method steps of recognizing and learning patterns in an adaptive learning network, comprising:

receiving an input pattern made up of one or more basic components, the one or more basic components measured in event time corresponding to an order of occurrence of the one or more basic components in the input pattern;

searching an adaptive learning network for a cel structure having a basic component of the one or more basic components starting in the order of occurrence in the input pattern;

positively reinforcing the cel structure having the basic component;

creating one or more cel structures to hold the one or more basic components and linking the created one or more cel structures to the adaptive learning network according to its event time, if not found.