Feature selection algorithm under conditions of noisy data and limited recording

Patent metadata

Field	Value
Publication number	US-10318892-B2
Application number	US-201615378153-A
Country	US
Kind code	B2
Filing date	Dec 14, 2016
Priority date	Jun 30, 2015
Publication date	Jun 11, 2019
Grant date	Jun 11, 2019

Abstract

Application of inter-class and intra-class filtering, based on aggregate point-to-point distances, to vector data for purposes of filtering the vector data for purposes of pattern recognition. In some embodiments: (i) the inter-class filtering is based on Euclidean distance, in all dimensions, between vector data points in vector space; and/or (ii) the intra-class filtering is based on a distance, in all dimensions, between vector data points in vector space.

First claim

What is claimed is:

1. A method comprising:
(I) generating a set of filtered vector data by:
applying, by machine logic of the computer, an inter-class filtering to a set of vector data, wherein:
the inter-class filtering reduces a number of vector values under consideration such that signal overlap is reduced between at least two classes of a plurality of classes,
each class of the plurality of classes respectively (i) represents a source for vector data and (ii) is associated with an entity or an object,
the set of vector data includes a plurality of vector values from each class of the plurality of classes, and
the inter-class filtering is based, at least in part, on an inter-class distance, wherein the inter-class distance is based on a sum of distances between:
a subject vector value, of a given class, in the plurality of vector values; and
at least some of the vector values in the plurality of vector values of at least one other class of the plurality of classes; and

applying, by machine logic of the computer, an intra-class filtering to the set of vector data, wherein the intra-class filtering is based, at least in part, on an intra-class distance; and

(II) responsive to a determination that a pattern in the set of filtered vector data matches a pattern assigned to a given class, executing an action that is associated with the given class.

2. The method of claim 1 , the method further comprising:
determining that the set of filtered vector data matches the pattern assigned to a given class based on a comparison of retained vectors for a plurality of classes to the pattern in the set of filtered vector data; and
selecting the action associated with the given class based on the matching, wherein the action associated with the given class includes one or more of: (i) creating a new pattern recognition signature; (ii) refining an existing pattern recognition signature; (iii) discarding an ineffective pattern recognition signature; (iv) assigning a given pattern to a given class based on a set of retained vectors; and (v) performing additional pattern matching, and (vi) determining a class to which a given set of vectors belongs based on a previously performed action that is associated with the given class.

3. The method of claim 1 , wherein the intra-class distance is based on a sum of distances between:
a subject vector value in the plurality of vector values in a first class of the plurality of classes; and
at least some of the vector values in the plurality of vector values of the first class.

4. The method of claim 1 , further comprising:
performing, by machine logic of the computer, a pattern recognition related action, wherein the pattern recognition related action is based, at least in part, on the inter-class filtering and the intra-class filtering.

5. The method of claim 4 , wherein performing a pattern recognition related action includes:
assigning, by machine logic of the computer, a pattern to each class in the plurality of classes, wherein the pattern is based on: the inter-class filtering or the intra-class filtering.

6. The method of claim 4 , wherein performing a pattern recognition related action includes:
determining, by machine logic of the computer, a class to which an unknown vector value in the plurality of vector values belongs based, at least in part, on a previous pattern recognition related action.

7. The method of claim 1 , wherein:
the inter-class distance is based on a sum of Euclidean distances between: (i) a subject vector value in the plurality of vector values; and (ii) at least some of the vector values in the plurality of vector values of at least one other class of the plurality of classes; and
the intra-class distance is based on a sum of Euclidean distances between: (i) a subject vector value in the plurality of vector values in a first class of the plurality of classes; and (ii) at least some of the vector values in the plurality of vector values of the first class.

8. A computer program product comprising:
a computer readable storage medium having stored thereon:
first instructions executable by a device to cause the device to apply, by machine logic of the computer, an inter-class filtering to a set of vector data, wherein:
the inter-class filtering reduces a number of vector values under consideration such that signal overlap is reduced between at least two classes of a plurality of classes,
each class of the plurality of classes respectively (i) represents a source for vector data and (ii) is associated with an entity or an object,
the set of vector data includes a plurality of vector values from each class of the plurality of classes, and
the inter-class filtering is based, at least in part, on an inter-class distance wherein the inter-class distance is based on a sum of distances between:
a subject vector value, of a given class, in the plurality of vector values; and
at least some of the vector values in the plurality of vector values of at least one other class of the plurality of classes; and

second instructions executable by a device to cause the device to apply, by machine logic of the computer, an intra-class filtering to the set of vector data, wherein the intra-class filtering is based, at least in part, on an intra-class distance; and
third instructions executable by a device to cause the device to respond to a determination that a pattern in the set of filtered vector data matches a pattern assigned to a given class by executing an action that is associated with the given class.

9. The computer program product of claim 8 , wherein the further instructions further comprise instructions to:
determine that the set of filtered vector data matches the pattern assigned to a given class based on a comparison of retained vectors for a plurality of classes to the pattern in the set of filtered vector data; and
select the action associated with the given class based on the matching, wherein the action associated with the given class includes one or more of: (i) creating a new pattern recognition signature; (ii) refining an existing pattern recognition signature; (iii) discarding an ineffective pattern recognition signature; (iv) assigning a given pattern to a given class based on a set of retained vectors; and (v) performing additional pattern matching, and (vi) determining a class to which a given set of vectors belongs based on a previously performed action that is associated with the given class.

10. The computer program product of claim 8 , wherein the intra-class distance is based on a sum of distances between:
a subject vector value in the plurality of vector values in a first class of the plurality of classes; and
at least some of the vector values in the plurality of vector values of the first class.

11. The computer program product of claim 8 , further comprising:
fourth instructions executable by a device to cause the device to perform, by machine logic of the computer, a pattern recognition related action, wherein the pattern recognition related action is based, at least in part, on the inter-class filtering and the intra-class filtering.

12. The computer program product of claim 11 , wherein third instructions to perform a pattern recognition related action include:
fifth instructions executable by a device to cause the device to assign, by machine logic of the computer, a pattern to each class in the plurality of classes, wherein the pattern is based on: the inter-class filtering or the intra-class filtering.

13. The computer program product of claim 11 , wherein third instructions to perform a pattern recognition related action include:
six

Assignees

• IBM

Inventors

• Arunkumar Saritha
• Yang Su

Classifications

• G06F2218/08

  Feature extraction · CPC title

• G06F18/2115

  by evaluating different subsets according to an optimisation criterion, e.g. class separability, forward selection or backward elimination · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06F18/24137

  Distances to cluster centroïds · CPC title

• G06F18/241

  relating to the classification model, e.g. parametric or non-parametric approaches · CPC title