Computer-implemented semi-supervised learning systems and methods

Abstract

Computer-implemented systems and methods for determining a subset of unknown targets to investigate. For example, a method can be configured to receive a target data set, wherein the target data set includes known targets and unknown targets. A supervised model such as a neural network model is generated using the known targets. The unknown targets are used with the neural network model to generate values for the unknown targets. Analysis with an unsupervised model is performed using the target data set in order to determine which of the unknown targets are outliers. A comparison of list of outlier unknown targets is performed with the values for the unknown targets that were generated by the neural network model. The subset of unknown targets to investigate is determined based upon the comparison.

It is claimed: 
     
         1 . A processor-implemented method for determining a subset of unknown targets to investigate, comprising:
 receiving a target data set;   wherein the target data set includes known targets and unknown targets;   generating a neural network model using the known targets;   using the unknown targets with the neural network model to generate values for the unknown targets;   performing outlier detection analysis using the target data set in order to determine which of the unknown targets are outliers;   performing a comparison of list of outlier unknown targets with the values for the unknown targets that were generated by the neural network model;   wherein the subset of unknown targets to investigate is determined based upon the comparison.   
     
     
         2 . The method of  claim 1 , wherein the target data set is indicative of entities to investigate with respect to fraud. 
     
     
         3 . The method of  claim 2 , wherein the fraud includes tax evasion fraud or purchase card fraud. 
     
     
         4 . The method of  claim 2 , wherein the values for the unknown targets that were generated by the neural network model are scores indicative of whether fraud has occurred. 
     
     
         5 . The method of  claim 1 , wherein the performing of the outlier detection analysis includes compressing at least a portion the unknown targets into a lower dimensional representation and uncompressing intermediate results of the compressing process into a higher dimensional representation;
 determining whether an unknown target is an outlier based error that exists between the uncompressed form and the unknown target.   
     
     
         6 . The method of  claim 5 , wherein said compressing is performed by using a compression neural network approach. 
     
     
         7 . The method of  claim 1 , wherein the performing of the outlier detection analysis includes using a nonlinear replicator neural network approach to determine which of the unknown targets are outliers. 
     
     
         8 . The method of  claim 1 , wherein the list of outlier unknown targets is a rank order list of the unknown targets that was generated through use of a compression neural network approach. 
     
     
         9 . The method of  claim 1 , wherein the performing of the comparison includes comparing based upon a pre-specified criterion or criteria the list of outlier unknown targets with the values for the unknown targets that were generated by the neural network model. 
     
     
         10 . The method of  claim 9 , wherein the pre-specified criterion or criteria contain criterion or criteria for including in the subset of unknown targets to investigate those unknown targets that have a value that satisfies a value scoring threshold and that have been determined to be outliers by the outlier detection analysis. 
     
     
         11 . The method of  claim 1 , wherein targets in the determined subset of unknown targets are investigated;
 wherein the targets in the determined subset become known because of the investigation.   
     
     
         12 . The method of  claim 11  further comprising:
 retraining the neural network using the known targets and the targets that have become known because of the investigation. 
 
     
     
         13 . The method of  claim 12  further comprising:
 using the retrained neural network upon the unknown targets in order to generate values for the unknown targets. 
 
     
     
         14 . The method of  claim 13  further comprising:
 performing a second outlier detection analysis in order to determine a second list of which of the unknown targets are outliers; 
 performing a second comparison of the second list of outlier unknown targets with the values for the unknown targets that were generated by the retrained neural network model; 
 wherein a second subset of unknown targets to investigate is determined based upon the second comparison. 
 
     
     
         15 . The method of  claim 1  further comprising:
 combining values for the unknown targets that were generated by the neural network model with outlier results generated by the outlier detection process in order to optimize the overall model performance. 
 
     
     
         16 . The method of  claim 1 , wherein an unknown target being identified as an outlier by the outlier detection process is an indication of anomalous activity. 
     
     
         17 . The method of  claim 16 , wherein an unknown target being identified as an outlier by the outlier detection process is an indication of fraudulent activity. 
     
     
         18 . A data signal that is transmitted using a network, wherein the data signal includes the subset of unknown targets to investigate of  claim 1 ;
 wherein the data signal comprises packetized data that is transmitted across the network.   
     
     
         19 . A processor-implemented system for determining a subset of unknown targets to investigate, comprising:
 a data store to store a target data set;   wherein the target data set includes known targets and unknown targets;   model generation software instructions configured to generate a neural network model using the known targets;   wherein the unknown targets are used with the neural network model to generate values for the unknown targets;   software instructions to perform outlier detection analysis using the target data set in order to determine which of the unknown targets are outliers;   comparison software instructions configured to perform a comparison of list of outlier unknown targets with the values for the unknown targets that were generated by the neural network model;   wherein the subset of unknown targets to investigate is determined based upon the comparison.   
     
     
         20 . Computer software stored on one or more computer readable mediums, the computer software comprising program code for determining a subset of unknown targets to investigate, said method comprising:
 receiving a target data set;   wherein the target data set includes known targets and unknown targets;   generating a neural network model using the known targets;   using the unknown targets with the neural network model to generate values for the unknown targets;   performing outlier detection analysis using the target data set in order to determine which of the unknown targets are outliers;   performing a comparison of list of outlier unknown targets with the values for the unknown targets that were generated by the neural network model;   wherein the subset of unknown targets to investigate is determined based upon the comparison.