Bayesian network based hybrid machine learning

Patent metadata

Field	Value
Publication number	US-10817779-B2
Application number	US-201715690672-A
Country	US
Kind code	B2
Filing date	Aug 30, 2017
Priority date	Aug 30, 2017
Publication date	Oct 27, 2020
Grant date	Oct 27, 2020

Abstract

Data includes data with labels and data without labels. For data without labels a fuzzy rules system assigns pseudo labels. A computer processes the data with labels using a first cognitive neural network; processes the data with pseudo labels using a second cognitive neural network; and produces system outcomes by combining the results of the first and second cognitive neural networks. The computer obtains feedback on the system outcomes, and modifies parameters of the fuzzy rule system in response to the feedback.

First claim

What is claimed is:

1. A method of constructing a Bayesian network structure that combines a first neural network configured for supervised learning with a second neural network configured for supervised learning, the method comprising:
initializing the Bayesian network structure in a memory of a computer system;
receiving, by the computer system, electronic data that includes data with labels and data without labels;
for the data without labels, assigning pseudo labels, wherein a fuzzy rules system assigns the pseudo labels;
establishing weights of the first neural network by processing the data with labels using the first neural network;
establishing weights of the second neural network by processing the data with pseudo labels using the second cognitive neural network;
producing system outcomes by combining results of the first and second neural networks;
obtaining, by the computer system, feedback on the system outcomes; and
updating the Bayesian network structure in response to the feedback by the second neural network modifying parameters of the fuzzy rule system.

2. The method of claim 1 wherein the pseudo labels are assigned in response to fuzzy rules.

3. The method of claim 2 wherein the fuzzy rules assess probability of fraud in response to identifying statistical outliers in the data.

4. The method of claim 2 wherein the fuzzy rules are developed in response to domain expert knowledge.

5. The method of claim 4 wherein the second cognitive neural network updates sensitivity and confidence parameters of the fuzzy rules in response to the feedback.

6. The method of claim 1 wherein the pseudo labels are assigned in response to fuzzy rules and in response to group risk scenarios.

7. The method of claim 6 wherein the group risk scenarios are developed in response to domain expert knowledge.

8. A non-transitory computer readable medium embodying computer executable instructions which when executed by a computer cause the computer to facilitate the method of:
initializing in the computer a Bayesian network structure that combines a first neural network configured for supervised learning with a second neural network configured for supervised learning;
receiving, by the computer, data that includes data with labels and data without labels;
for the data without labels, assigning pseudo labels, wherein a fuzzy rules system assigns the pseudo labels;
establishing weights of the first neural network by processing the data with labels using the first neural network;
establishing weights of the second neural network by processing the data with pseudo labels using the second neural network;
producing system outcomes by combining the results of the first and second neural networks;
obtaining, by the computer, feedback on the system outcomes; and
updating the Bayesian network structure in response to the feedback by the second neural network modifying parameters of the fuzzy rule system.

9. The medium of claim 8 wherein the pseudo labels are assigned in response to fuzzy rules.

10. The medium of claim 9 wherein the fuzzy rules assess probability of fraud in response to identifying statistical outliers in insurance claim data.

11. The medium of claim 9 wherein the fuzzy rules are developed in response to domain expert knowledge.

12. The medium of claim 11 wherein the second cognitive neural network updates sensitivity and confidence parameters of the fuzzy rules in response to the feedback.

13. The medium of claim 8 wherein the pseudo labels are assigned in response to fuzzy rules and in response to group risk scenarios.

14. The medium of claim 13 wherein the group risk scenarios are developed in response to domain expert knowledge.

15. An apparatus comprising:
a memory embodying computer executable instructions; and
at least one processor, coupled to the memory, and operative by the computer executable instructions to implement a method of:
initializing in the memory a Bayesian network structure that combines a first neural network configured for supervised learning with a second neural network configured for supervised learning;
receiving, by the processor, data that includes data with labels and data without labels, the labels indicating fraud or not fraud;
for the data without labels, assigning pseudo labels, wherein a fuzzy rules system assigns the pseudo labels;
establishing weights of the first neural network by processing the data with labels using the first neural network;
establishing weights of the second neural network by processing the data with pseudo labels using the second neural network;
producing system outcomes by combining the results of the first and second neural networks;
obtaining, by the processor, feedback on the system outcomes; and
updating the Bayesian network structure in response to the feedback by the second neural network modifying parameters of the fuzzy rule system.

16. The apparatus of claim 15 wherein the pseudo labels are assigned in response to fuzzy rules.

17. The apparatus of claim 16 wherein the fuzzy rules assess probability of fraud in response to identifying statistical outliers in insurance claim data.

18. The apparatus of claim 17 wherein the fuzzy rules are developed in response to domain expert knowledge.

19. The apparatus of claim 17 wherein the second cognitive neural network updates sensitivity and confidence parameters of the fuzzy rules in response to the feedback.

20. The apparatus of claim 15 wherein the pseudo labels are assigned in response to fuzzy rules and in response to group risk scenarios.

Assignees

• IBM

Inventors

• Bourgoin Elizabeth
• Buchholz Christopher A
• Kessler Eric M
• Song Liyang
• Zhu Zhihuai

Classifications

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N3/043

  based on fuzzy logic, fuzzy membership or fuzzy inference, e.g. adaptive neuro-fuzzy inference systems [ANFIS] · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/042

  Knowledge-based neural networks; Logical representations of neural networks · CPC title