Multi-variable pattern recognition for predictive deep learning models

Patent metadata

Field	Value
Publication number	US-12488561-B2
Application number	US-202016887522-A
Country	US
Kind code	B2
Filing date	May 29, 2020
Priority date	May 29, 2020
Publication date	Dec 2, 2025
Grant date	Dec 2, 2025

Abstract

Pattern recognition by receiving a set multi-variable data records, each record including a plurality of variables, representing at least two of the plurality of variables as geometric shapes, defining a boundary enclosing the geometric shapes, configuring at least one geometric shape to move within the boundary, capturing a location of each of the geometric shapes within the boundary as a system state, one or more times, combining one or more system states as a system signature, providing a model trained to recognize patterns in system signatures, and recognizing a pattern in the system signature.

First claim

What is claimed is:

1 . A computer implemented method for pattern recognition, the method comprising:
receiving, by one or more computer processors over a network, a set of multi-variable data records, each record comprising a plurality of variables; representing, by the one or more computer processors, at least two of the plurality of variables as geometric shapes; defining, by the one or more computer processors, a boundary enclosing the geometric shapes; initializing, by the one or more computer processors, motion of at least one geometric shape within the boundary; capturing, by the one or more computer processors, a location of each of the geometric shapes within the boundary as a system state, at a succession of instances; combining, by the one or more computer processors, one or more of the system states to form a first system signature; storing the system signature in persistent storage; providing, by the one or more computer processors, a model trained to recognize patterns in system signatures, wherein training the model comprises receiving labeled input data records, the data records comprising variable data associated with a state of a system, the label indicating an outcome; processing the labeled input data records by;
defining data particles, initial locations, movement parameters and boundaries for each record of the input data;
processing a series of input data records as a system;
moving and changing data particles according to changes in input data variable values;
collecting system states according to locations of the data particles;
forming system signatures according to the system states; and
using the system signatures, back propagation and a defined loss function to tune network node weights of the model to train the model to associate system signatures with the outcome;
storing the model in persistent storage; predicting, by the one or more computer processors, using the first system signature and the trained model, a system component failure; and initiating, by the one or more computer processors, over the network, a response to prevent the system component failure.

2 . The computer implemented method according to claim 1 , further comprising representing each variable of the set of multi-variable data records as a separate geometric shape.

3 . The computer implemented method according to claim 1 , further comprising representing multiple variables of the set of multi-variable data records as a single geometric shape.

4 . The computer implemented method according to claim 1 , wherein initializing at least one geometric shape to move within the boundary comprises configuring the geometric shape to move according to a magnitude of a variable of the set of multi-variable data records.

5 . The computer implemented method according to claim 1 , wherein the model is trained to classify patterns in the system signature.

6 . The computer implemented method according to claim 1 , wherein initializing at least one geometric shape to move within the boundary comprises configuring the geometric shape to move according to changes in values of the set of multi-variable data records over time.

7 . The computer implemented method according to claim 1 , wherein the boundary comprises more than two dimensions.

8 . A computer program product for pattern recognition, the computer program product comprising one or more computer readable storage devices and program instructions collectively stored on the one or more computer readable storage devices, the stored program instructions comprising:
program instructions to receive a set of multi-variable data records, each record comprising a plurality of variables, over a network; program instructions to represent at least two of the plurality of variables as geometric shapes; program instructions to define a boundary enclosing the geometric shapes; program instructions to initialize motion of at least one geometric shape within the boundary; program instructions to capture a location of each of the geometric shapes within the boundary as a system state, at a succession of instances; program instructions to combine one or more of the system states to form a first system signature; program instructions to store the first system signature in persistent storage; program instructions to provide a model trained to recognize patterns in system signatures; wherein training the model comprises receiving labeled input data records, the data records comprising variable data associated with a state of a system, the label indicating an outcome; processing the labeled input data records by;
defining data particles, initial locations, movement parameters and boundaries for each record of the input data;
processing a series of input data records as a system;
moving and changing data particles according to changes in input data variable values;
collecting system states according to locations of the data particles;
forming system signatures according to the system states; and
using the system signatures, back propagation and a defined loss function to tune network node weights of the model to train the model to associate system signatures with the outcome;
program instructions to store the model in persistent storage; program instructions to predict a system component failure using the first system signature and the model; and program instructions to initiate a response to prevent the system component failure.

9 . The computer program product according to claim 8 , the stored program instructions further comprising program instructions to represent each variable of the set of multi-variable data records as a separate geometric shape.

10 . The computer program product according to claim 8 , the stored program instructions further comprising program instructions to represent multiple variables of the set of multi-variable data records as a single geometric shape.

11 . The computer program product according to claim 8 , the stored program instruction to initialize motion of at least one geometric shape to move within the boundary comprising program instructions to configure the geometric shape to move according to a magnitude of a variable of the set of multi-variable data records.

12 . The computer program product according to claim 8 , wherein the model is trained to classify patterns in the system signature.

13 . The computer program product according to claim 8 , the stored program instructions to initialize the at least one geometric shape to move within the boundary comprising program instructions to configure the geometric shape to move according to changes in values of the set of multi-variable data records over time.

14 . The computer program product according to claim 8 , wherein the boundary comprises more than two dimensions.

15 . A computer system for pattern recognition, the computer system comprising:
one or more computer processors; one or more computer readable storage devices; and stored program instructions on the one or more computer readable storage devices for execution by the one or more computer processors, the stored program instructions comprising: program instructions to receive a set of multi-variable data records, each record comprising a plurality of variables, over a network; program instructions to represent at least two of the plurality of variables as geometric shapes; program instructions to define a boundary enclosing the geometric shapes; program instructions to initialize motion of at least one geometric shape within the boundary; program instructions to capt

Assignees

• IBM

Inventors

• Bacarella David
• Majumdar Arunava
• Roorda Stephan Joseph
• Patel Sumit

Classifications

• G06N20/00

  Machine learning · CPC title

• G06V10/776

  Validation; Performance evaluation · CPC title

• G06V10/774

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

• G06V10/7715

  Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods · CPC title

• G06F18/21375

  involving differential geometry, e.g. embedding of pattern manifold · CPC title