System and method for heterogeneous relational kernel learning

Patent metadata

Field	Value
Publication number	US-11354600-B2
Application number	US-201916536926-A
Country	US
Kind code	B2
Filing date	Aug 9, 2019
Priority date	Mar 5, 2019
Publication date	Jun 7, 2022
Grant date	Jun 7, 2022

Abstract

A computer-implemented method for generating an interpretable kernel embedding for heterogeneous data. The method can include identifying a set of base kernels in the heterogeneous data; and creating multiple sets of transformed kernels by applying a unique composition rule or a unique combination of multiple composition rules to the set of base kernels. The method can include fitting the multiple sets into a stochastic process model to generate fitting scores that respectively indicate a degree of the fitting for each of the multiple sets; storing the fitting scores in a matrix; and standardizing the matrix to generate the interpretable kernel embedding for the heterogeneous data.

First claim

What is claimed is:

1. A computer-implemented method for generating an interpretable kernel embedding for heterogeneous data, the method comprising:
obtaining, in a processing system, heterogeneous data including at least two or more subsets differing in linear or non-linear trends or behavior;
identifying, in a processing device of the processing system, an initial set of kernels in the heterogeneous data;
expanding, in the processing device of the processing system, the initial set of kernels by combining each kernel in the initial set with one or more other kernels in the initial set;
fitting, in the processing device of the processing system, each kernel in the expanded set of kernels using a stochastic process model to generate a fitting score that grades a degree in which a respective kernel can model at least one common feature among two or more of the subsets of the heterogeneous data;
storing, in a memory device of the processing system, the fitting scores in a matrix;
standardizing, via the processing device of the processing system, the matrix to generate the interpretable kernel embedding for an embedding space for the heterogeneous data including at least the two or more subsets having the at least one common feature; and
generating, via the processing device of the processing system, an output that identifies at least groupings of the heterogeneous data based on distances between the subsets in the interpretable kernel embedding.

2. The method of claim 1 , wherein the stochastic process model is a Gaussian process model.

3. The method of claim 1 , wherein the interpretable kernel embedding determines one or more groupings in the embedding space.

4. The method of claim 1 , wherein the interpretable kernel embedding detects one or more patterns in the embedding space.

5. The method of claim 1 , wherein the interpretable kernel embedding detects one or more anomalies in the embedding space.

6. The method of claim 1 , wherein each row of the standardized matrix corresponds to a representation of a linear or non-linear trend or behavior in the embedding space, and each column is a dimension of the embedding space, each column being associated with a specific kernel.

7. The method of claim 6 , wherein the linear or non-linear trend or behavior is a time series.

8. A system for generating an interpretable kernel embedding for heterogeneous data, the system comprising:
an interface configured to obtain heterogeneous data including at least two or more subsets differing in linear or non-linear trends or behavior;
a processing device configured to include:
an identification module configured to identify an initial set of kernels in the heterogeneous data;
a transformation module configured to expand the initial set of kernels by combining each kernel in the initial set with one or more other kernels in the initial set;
a scoring module configured to fit each kernel in the expanded set of kernels using a stochastic process model, and generate a fitting score that grades a degree in which a respective kernel can model at least one common feature among two or more of the subsets of the heterogeneous data; and
a normalization module configured to save the fitting scores in a matrix and standardize the matrix to generate an interpretable kernel embedding for an embedding space for the heterogeneous data including at least the two or more subsets having the at least one common feature,

the interface configured to generate an output that identifies at least groupings of the heterogeneous data based on distances between the subsets in the interpretable kernel embedding.

9. The system of claim 8 , wherein the stochastic process model is a Gaussian process model.

10. The system of claim 8 , wherein the interpretable kernel embedding determines one or more groupings in the embedding space.

11. The system of claim 8 , wherein the interpretable kernel embedding detects one or more patterns in the embedding space.

12. The system of claim 8 , wherein the interpretable kernel embedding detects one or more anomalies in the embedding space.

13. The system of claim 8 , wherein each row of the standardized matrix corresponds to a representation of a linear or non-linear trend or behavior in the embedding space, and each column is a dimension of the embedding space, each column being associated with a specific kernel.

14. The system of claim 13 , wherein the linear or non-linear trend or behavior is a time series.

Assignees

• Booz Allen Hamilton Inc

Inventors

• Nguyen Andre Tai
• Raff Edward

Classifications

• G06N7/01

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

• G06N20/10 Primary

  using kernel methods, e.g. support vector machines [SVM] · CPC title

• G06N20/20

  Ensemble learning · CPC title

• G06N7/005

  Physics · mapped topic