Data processing apparatus and method for automatically generating a classification component

Abstract

Data processing apparatus operative ( 100 ) to generate a classification component ( 106 ) is disclosed. The data processing apparatus is configured to provide a template classifier bank ( 1001 ) comprising a plurality of classifier modules ( 1004 ), each classifier module operative to receive training data comprising data elements having one of two or more known class affiliations and to output a class affiliation estimate value for each input data element. The data processing apparatus is further configured to derive a combination of the class affiliation estimate values providing a highest correlation to the two or more known class affiliations, and to generate a classification component formed of a resultant classifier bank comprising a combination of the plurality of classifier modules corresponding to the combination of estimate values providing the highest correlation.

1 .- 62 . (canceled) 
   
   
       63 . A data processing apparatus operative to generate a classification component, said data processing apparatus configured to provide a template classifier bank comprising a plurality of classifier modules, each classifier module operative to receive training data comprising data elements having one of two or more known class affiliations and to output a class affiliation estimate value for each input data element, said data processing apparatus further configured to derive a combination of said class affiliation estimate values providing a highest correlation to said two or more known class affiliations, and to generate a classification component formed of a resultant classifier bank comprising a combination of said plurality of classifier modules corresponding to said combination of estimate values providing said highest correlation. 
   
   
       64 . The data processing apparatus according to  claim 63 , wherein said resultant classifier bank comprises a sub-set of said plurality of classifier modules forming said template classifier bank. 
   
   
       65 . The data processing apparatus according to  claim 64 , configured to deselect classifier modules from said template classifier bank to generate said resultant classifier bank. 
   
   
       66 . The data processing apparatus according to  claim 64 , configured to apply a zero weighting to outputs of said template classifier bank not corresponding to said combination of estimate values providing said highest correlation. 
   
   
       67 . The data processing apparatus according to  claim 63 , further configured such that said combination of class affiliation estimate values provides respective highest correlation values to said two or more known class affiliations, and to generate said resultant classifier bank comprising a combination of said plurality of classifier modules corresponding to said combination of class affiliation estimate values providing said respective highest correlation values. 
   
   
       68 . The data processing apparatus according to  claim 63 , further configured such that said highest correlation is represented by a highest correlation value derived from a mathematical combination of said class affiliation estimate values, and said resultant classifier bank comprises a combination of said plurality of classifier modules corresponding to said combination of class affiliation estimate values providing said highest correlation value. 
   
   
       69 . The data processing apparatus according to  claim 68 , wherein said mathematical combination comprises a mean value of said class affiliation estimate values. 
   
   
       70 . The data processing apparatus according to  claim 63 , further configured to analyse said class affiliation estimate values using regression analysis in order to derive said combination of class affiliation estimate values. 
   
   
       71 . The data processing apparatus according to  claim 63 , further configured to input said class affiliation estimate values whose combination provided said respective highest correlation to each of said plurality of classifier modules of said template classifier bank to obtain further class affiliation estimate values and to derive a combination thereof which provides a further highest correlation to said two or more known class affiliations, and to generate a multi-stage classification component formed of said resultant classifier bank and a further resultant classifier bank comprising a combination of said plurality of classifier modules corresponding to said combination of further class affiliation estimate values providing said further highest correlation for said further highest correlation being greater than said highest correlation, and to arrange said multi-stage classification component such that class affiliation estimate values output from classifier modules of said resultant classifier bank are input to classifier modules of said further resultant classifier bank. 
   
   
       72 . The data processing apparatus according to  claim 71 , wherein said further resultant classifier bank comprises a sub-set of said plurality of classifier modules forming said template classifier bank. 
   
   
       73 . The data processing apparatus according to  claim 72 , configured to deselect classifier modules from said template classifier bank not corresponding to said combination of estimate values providing said highest correlation to generate said further resultant classifier bank. 
   
   
       74 . The data processing apparatus according to  claim 72 , configured to apply a zero weighting to outputs of said template classifier bank not corresponding to said combination of estimate values providing said highest correlation to generate said further resultant classifier bank. 
   
   
       75 . The data processing apparatus according to  claim 71 , further configured such that said combination of further class affiliation estimate values provides respective highest correlation values to said two or more known class affiliations, and to generate said further resultant classifier bank comprising a combination of said plurality of classifier modules corresponding to said combination of further class affiliation estimate values providing said respective highest correlation values. 
   
   
       76 . The data processing apparatus according to  claim 71 , further configured such that said highest correlation is represented by a highest correlation value derived from a mathematical combination of said further class affiliation estimate values, and said further resultant classifier bank comprises a combination of said plurality of classifier modules corresponding to said combination of further class affiliation estimate values providing said highest correlation value. 
   
   
       77 . The data processing apparatus according to  claim 76 , wherein said mathematical combination comprises a mean value of said further class affiliation estimate values. 
   
   
       78 . The data processing apparatus according to  claim 71 , further configured to analyse said further class affiliation estimate values using regression analysis in order to derive said combination of further class affiliation estimate values. 
   
   
       79 . The data processing apparatus according to  claim 63 , wherein said data elements are representative of source data of said known class affiliation. 
   
   
       80 . The data processing apparatus according to  claim 63 , further configured to normalise said training data. 
   
   
       81 . A data processing apparatus configured to:
 receive a multivariate data set comprising a plurality of data elements;   select from said data set a sub-set of data elements each data element of said sub-set having a deviation from a norm of said data set exceeding a threshold value; and   remove from said sub-set data elements having a contribution to the representation of said multivariate data set by said sub-set less than a second threshold value to form a final data set representative of said multivariate data set with a reduced number of sources of error.   
   
   
       82 . The data processing apparatus according to  claim 81 , further configured to derive said final data set by linearly combining data elements of said sub-set to provide an optimal representation of said multivariate data set. 
   
   
       83 . The data processing apparatus according to  claim 82 , wherein said optimal representation is in terms of the variance of the multivariate data set. 
   
   
       84 . The data processing apparatus according to  claim 82 , wherein said optimal representation is in terms of the uniformity of variance of the multivariate data set. 
   
   
       85 . The data processing apparatus according to  claim 81 , further configured to select said sub-set of data elements by:
 obtaining and ordering the Karhunen Loéve coordinate axes for said multivariate data set; and   selecting respective data elements for said subset from a n member row of said multivariate data set from positions in said row matrix corresponding to the column position of a maximum magnitude element of respective Karhunen Loéve coordinate axes vectors in order of greatest to least variance of the Karhunen Loéve coordinate axes vectors.   
   
   
       86 . The data processing apparatus, according to  claim 81 , further configured to derive said final set by applying a second Karhunen Loéve expansion to said sub-set. 
   
   
       87 . The data processing apparatus according to  claim 81 , configured to normalise data elements of said multivariate data set. 
   
   
       88 . The data processing apparatus according to  claim 81 , wherein said deviation comprises a deviation based on a least mean square estimate. 
   
   
       89 . The data processing apparatus according to  claim 81 , wherein said deviation comprises the variance of a respective data element. 
   
   
       90 . The data processing apparatus operative as a classification component and configured to:
 provide a first classifier bank comprising a plurality of classifier modules arranged to receive data elements representative of multivariate data;   provide a second classifier bank comprising a second plurality of classifier modules including at least one classifier module which is the same as a classifier module included in said first classifier bank; and   input class affiliation estimate values output from said first classifier bank to said second classifier bank.   
   
   
       91 . The data processing apparatus according to  claim 90 , wherein said first and second classifier bank are derived from a common classifier bank template. 
   
   
       92 . The data processing apparatus according to  claim 90 , further configured for providing said final data set representative of said multivariate data set to said first classifier bank. 
   
   
       93 . The data processing apparatus according to  claim 90 , further configured for providing a final data set representative of a multivariate data set output from said first classifier bank to said second classifier bank. 
   
   
       94 . A method of operating data processing apparatus to generate a classification component, comprising:
 providing a template classifier bank comprising a plurality of classifier modules, each classifier module operative to receive training data comprising data elements having one of two or more known class affiliations and to output a class affiliation estimate value for each input data element;   deriving a combination of said class affiliation estimate values providing a highest correlation to said two or more known class affiliations; and   generating a classification component formed of a resultant classifier bank comprising a combination of said plurality of classifier modules corresponding to said combination of estimate values providing said highest correlation.   
   
   
       95 . The method according to  claim 94 , wherein said resultant classifier bank comprises a sub-set of said plurality of classifier modules forming said template classifier bank. 
   
   
       96 . The method according to  claim 95 , further comprising deselecting classifier modules from said template classifier bank to generate said resultant classifier bank. 
   
   
       97 . The method according to  claim 95 , further comprising applying a zero weighting to outputs of said template classifier bank not corresponding to said combination of estimate values providing said highest correlation. 
   
   
       98 . The method according to  claim 94 , wherein said combination of class affiliation estimate values provides respective highest correlation values to said two or more known class affiliations, and further comprising generating said resultant classifier bank comprising a combination of said plurality of classifier modules corresponding to said combination of class affiliation estimate values providing said respective highest correlation values. 
   
   
       99 . The method according to  claim 94 , wherein said highest correlation is represented by a highest correlation value derived from a mathematical combination of said class affiliation estimate values, and said resultant classifier bank comprises a combination of said plurality of classifier modules corresponding to said combination of class affiliation estimate values providing said highest correlation value. 
   
   
       100 . The method according to  claim 99 , wherein said mathematical combination comprises a mean value of said class affiliation estimate values. 
   
   
       101 . The method according to  claim 94 , further configured to analyse said class affiliation estimate values using regression analysis in order to derive said combination of class affiliation estimate values. 
   
   
       102 . The method according to  claim 94 , further comprising inputting said class affiliation estimate values whose combination provided said respective highest correlation to each of said plurality of classifier modules of said template classifier bank to obtain further class affiliation estimate values, deriving a combination thereof which provides a further highest correlation to said two or more known class affiliations, generating a multi-stage classification component formed of said resultant classifier bank and a further resultant classifier bank comprising a combination of said plurality of classifier modules corresponding to said combination of further class affiliation estimate values providing said further highest correlation for said further highest correlation being greater than said highest correlation, and arranging said multi-stage classification component such that class affiliation estimate values output from classifier modules of said resultant classifier bank are input to classifier modules of said further resultant classifier bank. 
   
   
       103 . The method according to  claim 102 , wherein said further resultant classifier bank comprises a sub-set of said plurality of classifier modules forming said template classifier bank. 
   
   
       104 . The method according to  claim 103 , further comprising deselecting classifier modules from said template classifier bank not corresponding to said combination of estimate values providing said highest correlation to generate said further resultant classifier bank. 
   
   
       105 . The method according to  claim 103 , further comprising applying a zero weighting to outputs of said template classifier bank not corresponding to said combination of estimate values providing said highest correlation to generate said further resultant classifier bank. 
   
   
       106 . The method according to  claim 102 , wherein said combination of further class affiliation estimate values provides respective highest correlation values to said two or more known class affiliations, and to generate said further resultant classifier bank comprising a combination of said plurality of classifier modules corresponding to said combination of further class affiliation estimate values providing said respective highest correlation values. 
   
   
       107 . The method according to  claim 102 , wherein said highest correlation is represented by a highest correlation value derived from a mathematical combination of said further class affiliation estimate values, and said further resultant classifier bank comprises a combination of said plurality of classifier modules corresponding to said combination of further class affiliation estimate values providing said highest correlation value. 
   
   
       108 . The method according to  claim 107 , wherein said mathematical combination comprises a mean value of said further class affiliation estimate values. 
   
   
       109 . The method according to  claim 102 , further configured to analyse said further class affiliation estimate values using regression analysis in order to derive said combination of further class affiliation estimate values. 
   
   
       110 . The method according to  claim 94 , wherein said data elements are representative of source data of said known class affiliation. 
   
   
       111 . The method according to  claim 94 , further configured to normalise said training data. 
   
   
       112 . A method of operating data processing apparatus, comprising:
 receiving a multivariate data set comprising a plurality of data elements;   selecting from said data set a sub-set of data elements each data element of said sub-set having a deviation from a norm of said data set exceeding a threshold value; and   removing from said sub-set data elements having a contribution to the representation of said multivariate data set by said sub-set less than a second threshold value to form a final data set representative of said multivariate data set with a reduced number of sources of error.   
   
   
       113 . The method according to  claim 112 , further comprising deriving said final data set by linearly combining data elements of said sub-set to provide an optimal representation of said multivariate data set. 
   
   
       114 . The method according to  claim 113 , wherein said optimal representation is in terms of the variance of the multivariate data set. 
   
   
       115 . The method according to  claim 113 , wherein said optimal representation is in terms of the uniformity of variance of the multivariate data set. 
   
   
       116 . The method according to  claim 112 , further comprising selecting said sub-set of data elements by:
 obtaining and ordering the Karhunen Loéve coordinate axes for said multivariate data set; and   selecting respective data elements for said sub-set from a n member row of said multivariate data set from positions in said row matrix corresponding to the column position of a maximum magnitude element of respective Karhunen Loéve coordinate axes vectors in order of greatest to least variance of the Karhunen Loéve coordinate axes vectors.   
   
   
       117 . The method according to  claim 112 , further comprising deriving said final set by applying a second Karhunen Loéve expansion to said sub-set. 
   
   
       118 . The method according to  claim 112 , further comprising normalising data elements of said multivariate data set. 
   
   
       119 . The method according to  claim 112 , wherein said deviation comprises a deviation based on a least mean square estimate. 
   
   
       120 . The method according to  claim 112 , wherein said deviation comprises the variance of a respective data element. 
   
   
       121 . A method of operating data processing apparatus, comprising:
 providing a first classifier bank comprising a plurality of classifier modules arranged to receive data elements representative of multivariate data;   providing a second classifier bank comprising a second plurality of classifier modules including at least one classifier module which is the same as a classifier module included in said first classifier bank; and   inputting class affiliation estimate values output from said first classifier bank to said second classifier bank.   
   
   
       122 . The method according to  claim 121 , wherein said first and second classifier bank are derived from a common classifier bank template. 
   
   
       123 . The method according to  claim 121 , further configured for providing said final data set representative of said multivariate data set to said first classifier bank. 
   
   
       124 . The method according to  claim 121 , further configured for providing a final data set representative of a multivariate data set output from said first classifier bank to said second classifier bank. 
   
   
       125 . A data processing apparatus comprising:
 a receive module configured to receive a multivariate data set comprising a plurality of data elements;   a select module configured to select from said data set a sub-set of data elements each data element of said sub-set having a deviation from a norm of said data set exceeding a threshold value; and   a remove module configured to remove from said sub-set data elements having a contribution to the representation of said multivariate data set by said sub-set less than a second threshold value to form a final data set representative of said multivariate data set with a reduced number of sources of error