Energy minimization for classification, pattern recognition, sensor fusion, data compression, network reconstruction and signal processing

• US 7,174,048 B2
• Filed: 03/31/2005
• Issued: 02/06/2007
• Est. Priority Date: 12/29/1997
• Status: Expired due to Term

• Current Assignee: Abel Wolman, Jeff B. Glickman
• Original Assignee: Abel Wolman, Jeff B. Glickman
• Inventors: Glickman, Jeff B., Wolman, Abel
• Primary Examiner(s): DESIRE, GREGORY M
• Application Number:
  US11/096,126
• Publication Number:
  US 20050180638A1
• Time in Patent Office: 677 Days
• Field of Search: 382/133, 382/159, 382/170, 382/224, 382/260, 707/102
• US Class Current: 382/260
• CPC Class Codes: G06F 18/2137 based on criteria of topology preservation, e.g. multidimensional scaling or self-organising maps
• CPC Class Codes: Y10S 707/99932 Access augmentation or optimizing
• CPC Class Codes: Y10S 707/99942 Manipulating data structure, e.g. compression, compaction, compilation

Abstract

A data analyzer/classifier comprises using a preprocessing step, and energy minimization step, and a postprocessing step to analyze/classify data.

Claims

1. A classifier process for data, the classifier process comprising:

using individual differences multidimensional scaling with two or more input symmetric matrices into which the data to be classified has been converted to produce at least a source space output; and

using the source space output for at least one of to classify the data;

for pattern recognition;

for sensor fusion;

for optical character recognition; and

for data compression.

2. The classifier process of claim 1 wherein the step of using the source space comprises:

using the source space to classify the data, including a searching process selected from the group consisting of searching for clustering, searching for hyperplane discriminators, searching for decision surfaces and searching for classifying structures.

3. A method for data compression, the method comprising:

producing the two or more proximity matrices including the data to be compressed;

using individual differences multidimensional scaling upon the one or more input proximity matrices to produce a source space output and a common space output; and

using the source space output and the common space output as a compressed representation of the data.

4. A method for classifying data, the method comprising:

receiving input data for classification;

forming two or more symmetric matrices from the received input data;

applying individual differences multidimensional scaling to the two or more symmetric matrices to produce at least a source space output; and

using the source space output to classify the received data.

5. The method for classifying data of claim 4 wherein the step of forming two or more symmetric matrices from the received data comprises a processing step selected from the group consisting of forming two or more hollow symmetric matrices, entry-wise substituting the received data to populate two or more symmetric matrices, forming two or more proximity matrices, forming two or more distance matrices, and forming two or more Euclidean distance matrices.

6. The method for classifying data of claim 4 wherein the two or more symmetric matrices contain missing values.

7. The method for classifying data of claim 4 wherein the step of applying individual differences multidimensional scaling comprises applying nonmetric individual differences multidimensional scaling.

8. The method for classifying data of claim 4 wherein the step of using the source space output to classify the received data further comprises the steps of:

interpreting the source space output as containing contextual signatures of the received data; and

using the contextual signatures to classify the received data.

9. A pattern recognition method comprising:

receiving input data;

forming two or more symmetric matrices from the received input data;

applying individual differences multidimensional scaling to the two or more symmetric matrices to produce at least a source space output; and

using the source space output for pattern recognition.

10. The method for pattern recognition of claim 9 wherein the step of forming the two or more symmetric matrices from the received data comprises a processing step selected from the group consisting of forming two or more hollow symmetric matrices, entry-wise substituting the received data to populate two or more symmetric matrices, forming two or more proximity matrices, forming two or more distance matrices and forming two or more Euclidean distance matrices.

11. The pattern recognition method of claim 9 wherein the two or more symmetric matrices contain missing values.

12. The pattern recognition method of claim 9 wherein the step of applying individual differences multidimensional scaling comprises applying nonmetric individual differences multidimensional scaling.

13. The pattern recognition method of claim 9 wherein the step of using the source space output to recognize patterns in the received data further comprises the steps of:

interpreting the source space output as containing contextual signatures of the received data; and

using the contextual signatures to recognize patterns in the received data.

14. A method for optical character recognition, the method comprising:

receiving input data representative of optimal character data;

forming two or more symmetric matrices using the received input data;

applying individual differences multidimensional scaling to the two or more symmetric matrices to produce at least a source space output; and

using the source space output for optical character recognition.

15. The method for optical character recognition of claim 14 wherein the step of forming the two or more symmetric matrices from the received data comprises a processing step selected from the group consisting of:

forming two or more hollow symmetric matrices, entry-wise substituting the received data to populate two or more symmetric matrices, forming two or more proximity matrices, forming two or more distance matrices and forming two or more Euclidean distance matrices.

16. The method for optical character recognition of claim 14 wherein the two or more symmetric matrices contain missing values.

17. The method for optical character recognition of claim 14 wherein the step of applying individual differences multidimensional scaling comprises applying nonmetric individual differences multidimensional scaling.

18. The method for optical character recognition of claim 14 wherein the step of using the source space output to recognize characters in the received data further comprises the steps of:

interpreting the source space output as containing contextual signatures of the received data; and

using the contextual signatures to recognize characters in the received data.

19. A method for data compression comprising:

receiving input data for compression;

forming two or more symmetric matrices from the received input data;

applying individual differences multidimensional scaling to the two or more symmetric matrices to produce at least a source space output; and

using the source space output for data compression.

20. The method for data compression of claim 19 wherein the step of forming the two or more symmetric matrices from the received data comprises a processing step selected from the group consisting of:

forming two or more hollow symmetric matrices, entry-wise substituting the received data to populate two or more symmetric matrices, forming two or more proximity matrices, forming two or more distance matrices and forming two or more Euclidean distance matrices.

21. The method for data compression of claim 19 wherein the source space output comprises contextual signatures of the received data.

22. The method for data compression of claim 19 wherein the contextual signatures comprise compressed data.