Directed dynamic data analysis

• US 20020159641A1
• Filed: 03/13/2002
• Published: 10/31/2002
• Est. Priority Date: 03/14/2001
• Status: Abandoned Application

• Current Assignee: Battelle Memorial Institute
• Original Assignee: Battelle Memorial Institute
• Inventors: Wiberg, John D., Hetzler, Elizabeth G., Whitney, Paul D., Havre, Susan L., Perrine, Kenneth A.
• Application Number:
  US10/097,719
• Publication Number:
  US 20020159641A1
• Time in Patent Office: Days
• US Class Current: 382/219
• CPC Class Codes: G06F 18/211 Selection of the most significant subset of features
• CPC Class Codes: G06F 18/41 Interactive pattern learning with a human teacher

Abstract

Several approaches are provided for designing algorithms that allow for fast retrieval, classification, analysis or other processing of data, with minimal expert knowledge of the data being analyzed, and further, with minimal expert knowledge of the math and science involved in building classifications and performing other statistical data analysis. Further, methods of analyzing data are provided where the information being analyzed is not easily susceptible to quantitative description.

Claims

1. A computer system for performing directed dynamic analysis comprising:

at least one processor;

a storage device accessible by said processor arranged to store a data set comprising digital representations of a plurality of data objects;

an input device coupled to said processor configured to accept input from a user; and

,

software executable by said at least one processor for;

iteratively exploring said data set at the object level by a user until a predetermined stopping criterion is met by performing for each iteration;

selecting by said user of at least two data objects from said data set;

providing by said user of at least one object level assessment that describes a relationship between said at least two data objects;

automatically taking diverse common measurements of features from said data objects based upon said at least one object level assessment and without further input from said user;

automatically constructing at least one of a rotation and a reweighting of said features based upon said at least one object level assessment and without further input from said user;

applying at least one of said rotation and said reweighting across at least a subset of said data set, and,

outputting at least one of a clustering, a projecting, and a summarizing of said subset in a manner that reflects said at least one object level assessment.

2. The computer system for performing directed dynamic analysis according to claim 1, wherein said software is further configured for outputting at least one of a feature set defining said features and feature vectors organizing the measured features.

3. The computer system for performing directed dynamic analysis according to claim 1, wherein said rotation is automatically constructed by:

constructing a data matrix comprising said diverse common measurements of features on each of said data objects;

extracting a sub-matrix from said data matrix comprising ones of said diverse common measurements for which said relationship is to be explored;

constructing a selection matrix that defines said relationship;

performing a canonical correlation computation using said sub-matrix and said selection matrix; and

,

applying the rotations obtained from said canonical correlation computation across at least a portion of said data set.

4. A computer system for performing directed dynamic analysis comprising:

a storage device having a plurality of digital representations of data objects stored thereon;

an input device configured to accept input from a user;

a processor coupled to said storage device and said input device programmed to allow a user to iteratively explore said data set at the object level until a predetermined stopping criterion is met by executing for each iteration program code to;

allow said user to select at least two data objects from said data set;

receive as input from said user, at least one object level assessment that describes a relationship between said at least two data objects;

extract automatically, diverse common measurements of features from said data objects based upon said at least one object level assessment and without further input from said user;

construct automatically, at least one of a rotation and a reweighting of said features based upon said at least one object level assessment and without further input from said user;

apply at least one of said rotation and said reweighting across at least a subset of said data set, and,

output at least one of a cluster, a projection, and a summary of said subset in a manner that reflects said at least one object level assessment.

5. The computer system for performing directed dynamic analysis according to claim 4, wherein said processor further outputs at least one of a feature set defining said features and feature vectors organizing the measured features.

6. The computer system for performing directed dynamic analysis according to claim 4, wherein said processor is programmed to construct said rotation by executing code to:

construct a data matrix comprising said diverse common measurements of features on each of said data objects;

extract a sub-matrix from said data matrix comprising ones of said diverse common measurements for which said relationship is to be explored;

construct a selection matrix that defines said relationship;

perform a canonical correlation computation using said sub-matrix and said selection matrix; and

,

apply the rotations obtained from said canonical correlation computation across at least a portion of said data set.

7. A computer readable carrier including dynamic data analysis program code that causes a computer to perform operations comprising:

selecting by a user of at least two data objects from a data set;

providing by said user of at least one object level assessment that describes a relationship between said at least two data objects;

automatically taking diverse common measurements of features from said data objects based upon said at least one object level assessment and without further input from said user;

automatically constructing at least one of a rotation and a reweighting of said features based upon said at least one object level assessment and without further input from said user; and

,

applying at least one of said rotation and said reweighting across at least a subset of said data set.

8. The computer readable carrier according to claim 7, wherein said rotation is automatically constructed by:

constructing a data matrix comprising said diverse common measurements of features on each of said data objects;

extracting a sub-matrix from said data matrix comprising ones of said diverse common measurements for which said relationship is to be explored;

constructing a selection matrix that defines said relationship;

performing a canonical correlation computation using said sub-matrix and said selection matrix; and

,

applying the rotations obtained from said canonical correlation computation across at least a portion of said data set.

9. A system for performing a directed dynamic data analysis comprising:

a processor;

a storage device accessible by said processor arranged to store a data set comprising digital representations of a plurality of data objects;

an input device coupled to said processor configured to accept input from a user; and

,

software executable by said processor to derive a feature from said data set and to interact with a user for;

constructing a first matrix comprising select ones of said plurality of data objects, each having a plurality of measurements associated therewith;

constructing a second matrix describing the relationships between said select ones of said plurality of data objects in said first matrix;

applying a canonical correlations computation using said first and second matrices to derive at least one rotation; and

,

outputting said at least one rotation across at least a portion of said data set.

10. The system for performing a directed dynamic data analysis according to claim 9, wherein said software further uses the results of said canonical correlations computation to cluster and display visually at least a portion of said data set.

11. A system for performing a directed dynamic data analysis comprising:

a storage device comprising a data set having a plurality of digital representations of data objects stored thereon;

an input device configured to accept input from a user; and

,

a processor coupled to said storage device and said input device programmed to;

construct a first matrix comprising select ones of said plurality of data objects, each having a plurality of measurements associated therewith;

construct a second matrix describing the relationships between said select ones of said plurality of data objects in said first matrix;

apply a canonical correlations computation using said first and second matrices to derive at least one rotation; and

,

output said at least one rotation across at least a portion of said data set.

12. The system for performing a directed dynamic data analysis according to claim 11, wherein said processor further executes instructions to cluster and display visually the results of said canonical correlations applied to at least a portion of said data set.

15. The system for performing directed dynamic analysis according to claim 14, wherein each rule is selectively enabled and disabled, wherein said new projection of said images is defined by only applying enabled ones of each rule.

16. The system for performing directed dynamic analysis according to claim 14, wherein said projection is computed using a canonical correlation computation.

17. The system for performing directed dynamic analysis according to claim 14, wherein each rule defines an object level comparison.

18. The system for performing directed dynamic analysis according to claim 14, wherein said weight comprises a measure of similarity.

19. The system for performing directed dynamic analysis according to claim 18, wherein at least one setting of said weight comprises a neutral position.

13. A computer readable carrier including a directed dynamic analysis program code that causes a computer to perform operations comprising:

constructing a first matrix comprising select ones of a plurality of data objects stored on a computer storage device, each having a plurality of measurements associated therewith;

constructing a second matrix describing the relationships between said select ones of said plurality of data objects in said first matrix;

applying a canonical correlations computation using said first and second matrices to derive at least one rotation; and

,

outputting said at least one rotation across at least a portion of said data set.

14. A system for performing directed dynamic analysis comprising:

at least one processor;

a storage device accessible by said processor arranged to store a data set comprising digital representations of a plurality of data objects;

an input device coupled to said processor configured to accept input from a user; and

,

software executable by said at least one processor to derive a feature from said data set and to interact with a user for;

creating a window on a display device defining a workspace;

outputting images representing at least a portion of said data set in said workspace;

reiteratively analyzing said data set until a predetermined stopping criterion is met by performing at least one of;

defining at least one rule comprising;

selecting a subset of said data objects;

defining a condition that describes said data objects in said subset; and

,

assign a weight that established a measure of said condition;

outputting a new projection of said images by applying said at least one rule across said at least a portion of said data set, and,

modifying at least one rule.

20. A system for performing directed dynamic analysis comprising:

a storage device having a plurality of digital representations of data objects stored thereon;

an input device configured to accept input from a user; and

,

a processor coupled to said storage device and said input device programmed to;

create a window on a display device defining a workspace;

output images representing at least a portion of said data set in said workspace;

analyze reiteratively said data set until a predetermined stopping criterion is met by performing at least one to;

define at least one rule comprising;

select a subset of said data objects;

define a condition that describes said data objects in said subset; and

,

assign a weight that established a measure of said condition;

output a new projection of said images by applying said at least one rule across said at least a portion of said data set, and,

modify at least one rule.

21. The system for performing directed dynamic analysis according to claim 20, wherein each rule is selectively enabled and disabled, wherein said new projection of said images is defined by only applying enabled ones of each rule.

22. The system for performing directed dynamic analysis according to claim 20, wherein said projection is computed using a canonical correlation computation.

23. The system for performing directed dynamic analysis according to claim 20, wherein each rule defines an object level comparison.

24. The system for performing directed dynamic analysis according to claim 20, wherein said weight comprises a measure of similarity.

25. The system for performing directed dynamic analysis according to claim 20, wherein at least one setting of said weight comprises a neutral position.

28. The method of performing a directed dynamic analysis according to claim 27, wherein said rotation is applied across at least said subset of said data set by performing at least one of clustering, projecting, and summarizing said subset in a manner that reflects said at least one object level assessment.

29. The method of performing a directed dynamic analysis according to claim 27, wherein said reweightings are applied across at least said subset of said data, and the reweighted data are analyzed by performing at least one of clustering, projecting, and summarizing said subset in a manner that reflects said at least one object level assessment.

30. The method of performing a directed dynamic analysis according to claim 27, wherein said rotation defines a weight for each feature that is commensurate with said at least one object level assessment.

31. The method of performing a directed dynamic analysis according to claim 30, wherein each reweighting defines weights that are commensurate with said at least one object level assessment.

32. The method of performing a directed dynamic analysis according to claim 27, further comprising providing as an output, at least one of a feature set that defines said features, and feature vectors organizing the measured features.

33. The method of performing a directed dynamic analysis according to claim 27, wherein said rotation is automatically constructed by:

constructing a data matrix comprising said diverse common measurements of features on each of said data objects;

extracting a sub-matrix from said data matrix comprising ones of said diverse common measurements for which said relationship is to be explored;

constructing a selection matrix that defines said relationship;

performing a canonical correlation computation using said sub-matrix and said selection matrix; and

,

applying the rotations obtained from said canonical correlation computation across at least a portion of said data set.

35. The method of performing a directed dynamic data analysis on a data set having a plurality of data objects according to claim 34, wherein the rows of said data matrix correspond to said data objects and the columns of the data matrix correspond to the measurements of said data objects.

36. The method of performing a directed dynamic data analysis on a data set having a plurality of data objects according to claim 35, wherein said selection matrix comprises the same number of rows as said sub-matrix and the columns of said selection matrix correspond to the established relations.

37. The method of performing a directed dynamic data analysis on a data set having a plurality of data objects according to claim 34, wherein said relation comprise similarity and dissimilarity.

38. The method of performing a directed dynamic data analysis on a data set having a plurality of data objects according to claim 37, wherein similarity is assigned a value of one, and dissimilarity is assigned a value of zero.

39. The method of performing a directed dynamic data analysis on a data set having a plurality of data objects according to claim 34, wherein the rotations obtained from said canonical correlation computation are applied to said data set to create a visual clustering of said data set that reflects said relation.

26. A computer readable carrier including directed dynamic analysis program code that causes a computer to perform operations comprising:

creating a window on a display device defining a workspace;

outputting images representing at least a portion of a data set in said workspace;

reiteratively analyzing said data set until a predetermined stopping criterion is met by performing at least one of;

defining at least one rule comprising;

selecting a subset of said data objects;

defining a condition that describes said data objects in said subset; and

,

assign a weight that established a measure of said condition;

outputting a new projection of said images by applying said at least one rule across said at least a portion of said data set, and,

modifying at least one rule.

27. A method of performing a directed dynamic analysis of a data set having a plurality of data objects by expert and non-expert users comprising:

iteratively exploring said data set at the object level by a user until a predetermined stopping criterion is met by performing for each iteration;

selecting by a user of at least two data objects from said data set;

providing by said user of at least one object level assessment that describes a relationship between said at least two data objects;

automatically taking diverse common measurements of features from said data objects based upon said at least one object level assessment and without further input from said user;

automatically constructing at least one of a rotation and a reweighting of said features based upon said at least one object level assessment and without further input from said user; and

,

applying at least one of said rotation and said reweighting across at least a subset of said data set.

34. A method of performing a directed dynamic data analysis on a data set having a plurality of data objects comprising:

constructing a data matrix comprising a plurality of measurements taken on each of said data objects;

extracting a sub-matrix from said data matrix comprising ones of said plurality of measurements for which a relationship is to be explored;

constructing a selection matrix that defines said relationship;

performing a canonical correlation computation using said sub-matrix and said selection matrix; and

,

applying the rotations obtained from said canonical correlation computation across at least a portion of said data set.

40. A method of performing a directed dynamic data analysis on a data set having a plurality of data objects comprising:

constructing a first matrix comprising measurements associated with select ones of said plurality of data objects;

constructing a second matrix describing the relationships between said select ones of said plurality of data objects in said first matrix;

applying a canonical correlations computation using said first and second matrices to derive at least one rotation; and

,

applying said at least one rotation across at least a portion of said data set.

41. The method of performing a directed dynamic data analysis on a data set having a plurality of data objects according to claim 40, further comprising constructing a data matrix having a plurality of measurements associated therewith, corresponding to the portion of said data set to which said at least one rotation is applied.

42. The method of performing a directed dynamic data analysis on a data set having a plurality of data objects according to claim 40, further comprising visually displaying the obtained rotations from said canonical correlations operations across at least a portion of said data set to create a visual clustering that reflects the relationships described in said second matrix.

43. The method of performing a directed dynamic data analysis on a data set having a plurality of data objects according to claim 40, wherein each row of said first matrix defines an object of said data set and each column defines a measurement corresponding to an associates one of said objects.

44. The method of performing a directed dynamic data analysis on a data set having a plurality of data objects according to claim 40, wherein a select one of said elements in said first matrix is denoted by the variable a and satisfies the condition that α

_ij=λ

_j(O_i) is the jth measurement on the ith object.

45. The method of performing a directed dynamic data analysis on a data set having a plurality of data objects according to claim 40, wherein said select ones of said plurality of objects selected from said first matrix comprise data objects for which a similarity measure is desired.

46. The method of performing a directed dynamic data analysis on a data set having a plurality of data objects according to claim 40, wherein said select ones of said plurality of objects selected for said first matrix comprise data objects for which an object level comparison is desired.

47. The method of performing a directed dynamic data analysis on a data set having a plurality of data objects according to claim 40, wherein said second matrix comprises a number of rows corresponding to the number of rows in said first matrix and at least one column that corresponds to a rule describing the relationships between the rows in said first matrix.

49. The method of deriving a feature from a data set having a plurality of data objects according to claim 48, wherein the results of said canonical correlation analysis are clustered and displayed visually.

48. A method of deriving a feature from a data set having a plurality of data objects comprising:

constructing a data matrix A_nxm where α

_ij∈

R and α

_ij=λ

_j/(O_i) is the j^th measurement on the ith data object of said data set;

constructing a sub-matrix comprising at least two rows selected from said data matrix;

constructing a selection matrix comprising at least one column that establishes conditions that describe the relationship between said at least two rows in said sub-matrix;

applying a canonical correlation analysis to said sub-matrix and said selection matrix; and

,

applying the resulting rotations of said canonical correlation analysis across at least a subset of said data set.

50. A method of analyzing data objects comprising:

providing a data set having a plurality of data objects;

organizing said data set in a first projection;

selecting at least two data objects from said data set;

providing at least one object level comparison that describes a relationship between said at least two data objects;

extracting said at least one object level comparison across said data set; and

,

clustering said data set in a new projection that reflects the relationships described in said object level comparison.

51. A method of using a computer to perform directed dynamic analysis comprising:

projecting a data set in a workspace, said data set having a plurality of data objects;

selecting a subset of said data objects;

defining a rule that establishes a relationship between said data objects in said subset;

clustering said data set in said workspace based upon said rule; and

,

re-projecting the clustered version of said data set in said workspace.

52. The method of using a computer to perform directed dynamic analysis according to claim 51, further comprising assigning a weight to said subset of said data objects, wherein said rule establishes said relationship based at least in part upon said weight.

53. A method of using a computer to perform directed dynamic analysis comprising:

projecting a data set in a workspace, said data set having a plurality of data objects;

defining at least one rule comprising;

selecting a subset of said data objects; and

,

defining a condition that describes at least one relationship between said data objects in said subset;

defining a weight that establishes a measure of said condition; and

,

performing at least one of;

creating a new projection of said data set by applying the rule across said data set, and,

modifying at least one rule.

54. The method of using a computer to perform directed dynamic analysis according to claim 53, wherein modifying a select one of said at least one rule comprises at least one of changing said data subset, changing said condition, enabling said rule, disabling said rule, and removing said rule.

55. The method of using a computer to perform directed dynamic analysis according to claim 53, further comprising performing at least one of:

creating a new projection of said data set by applying the rule across said data set, and,

modifying at least one rule.

56. The method of using a computer to perform directed dynamic analysis according to claim 53, wherein modifying a select one of said at least one rule comprises at least one of changing said data subset, changing said condition, enabling said rule, disabling said rule, and removing said rule.