Robust pattern recognition system and method using socratic agents

• US 8,014,591 B2
• Filed: 09/13/2007
• Issued: 09/06/2011
• Est. Priority Date: 09/13/2006
• Status: Active Grant

• Current Assignee: Aurilab LLC
• Original Assignee: Aurilab LLC
• Inventors: Baker, James K.
• Primary Examiner(s): CARTER, AARON W
• Application Number:
  US11/898,636
• Publication Number:
  US 20080069437A1
• Time in Patent Office: 1,454 Days
• Field of Search: 382155-161, 700/47, 700/48, 704/231, 704/232, 706 45- 61
• US Class Current: 382/159
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06F 18/217 Validation; Performance evaluation; Active pattern learning techniques

Abstract

A computer-implemented pattern recognition method, system and program product, the method comprising in one embodiment: creating electronically a linkage between a plurality of models within a classifier module within a pattern recognition system such that any one of said plurality of models may be selected as an active model in a recognition process; creating electronically a null hypothesis between at least one model of said plurality of linked models and at least a second model among said plurality of linked models; accumulating electronically evidence to accept or reject said null hypothesis until sufficient evidence is accumulated to reject said null hypothesis in favor of one of said plurality of linked models or until a stopping criterion is met; and transmitting at least a portion of the electronically accumulated evidence or a summary thereof to accept or reject said null hypothesis to a pattern classifier module.

Claims

1. A computer-implemented pattern recognition method, comprising:

creating electronically a linkage between a plurality of models within a classifier module within a pattern recognition system such that any one of said plurality of models may be selected as an active model in a recognition process;

creating electronically a null hypothesis between at least one model of said plurality of linked models and at least a second model among said plurality of linked models;

accumulating electronically evidence to accept or reject said null hypothesis until sufficient evidence is accumulated to reject said null hypothesis in favor of one of said plurality of linked models or until a stopping criterion is met;

transmitting at least a portion of the electronically accumulated evidence or a summary thereof to accept or reject said null hypothesis to a pattern classifier module.

2. A pattern recognition method as in claim 1, further comprising:

subsequently performing recognition in which, when one null hypothesis is rejected in favor of a particular one of said plurality of linked models, said particular model is selected as the active model in said classifier module.

3. A pattern recognition method as in claim 1, further comprising:

obtaining a set of training data for training said classifier module;

obtaining a particular training sample for said classifier module and an associated label for said training sample;

creating a first model for said classifier module by training said classifier module on said set of training data not including said particular training sample;

creating a second model for said classifier module by training said classifier module on said set of training data including said particular training sample;

creating said linkage of said plurality of models in which said plurality of models includes at least said first model and said second model.

4. A pattern recognition method as in claim 3, further comprising:

annotating said particular training sample with the information obtained from said accumulating of evidence to accept or reject said null hypothesis.

5. A pattern recognition method as in claim 4, further comprising:

performing subsequent training skipping training samples and training with changed labels on the training samples in accord with the annotation obtained from said accumulation of evidence to accept or reject said null hypothesis.

6. A pattern recognition method as in claim 1, further comprising

obtaining a plurality of models resulting from different decisions at a decision point;

and

creating said linkage among the plurality of models resulting from the decision point.

7. A pattern recognition method as in claim 1, further comprising

obtaining a plurality of models differing from each other by having a differing number of elements in a given model data structure;

creating said linkage among the plurality of models having the differing number of elements in the given data structure;

creating electronically a null hypothesis between at least one model of said plurality of linked models and at least a second model among said plurality of linked models;

accumulating electronically evidence to accept or reject said null hypothesis until sufficient evidence is accumulated to reject said null hypothesis in favor of one of said plurality of linked models where the rejection criterion is based at least in part on a measure of the marginal cost for the differing number elements or until a stopping criterion is met;

transmitting at least a portion of the electronically accumulated evidence or a summary thereof to accept or reject said null hypothesis to a pattern classifier module.

8. A pattern recognition method as in claim 7, wherein the given data structure is a collection of lower-level models and the elements that differ in number are the lower-level models.

9. A pattern recognition method as in claim 8, further comprising creating at least one lower-level model by one-shot learning, and wherein the lower-level models differ in number at least in part due to the models created by one-shot learning.

10. A pattern recognition method as in claim 7, wherein the given data structure is a graphical structure and the elements that differ in number are arcs and nodes.