Data handling and machine learning

Patent metadata

Field	Value
Publication number	US-12217170-B2
Application number	US-202117216057-A
Country	US
Kind code	B2
Filing date	Mar 29, 2021
Priority date	Mar 27, 2020
Publication date	Feb 4, 2025
Grant date	Feb 4, 2025

Abstract

A method implemented by a software for a multimodal evaluation engine stored on a memory is provided herein. The software is executable by a processor coupled to the memory to cause the method. The method includes receiving multimodal signatures of an object of interest from inspection elements and processing the multimodal signatures to transform the multimodal signatures into formats. The method also includes generating data representations of the formats and detecting whether anomalies are present within the object of interest based on the data representations.

First claim

What is claimed is:

1. A method implemented by a software for a multimodal evaluation engine stored on a memory and executable by one or more processors coupled to the memory, the method comprising:
receiving, by the multimodal evaluation engine, a plurality of multimodal signatures of an object of interest from one or more inspection elements, wherein the plurality of multimodal signatures comprise visible light, infrared, electromagnetic interference, and laser profilometry;
processing, by the multimodal evaluation engine, the plurality of multimodal signatures to transform the plurality of multimodal signatures into one or more formats;
generating, by the multimodal evaluation engine, data representations of the one or more formats; and
detecting, by the multimodal evaluation engine, whether one or more anomalies are present within the object of interest based on the data representations.

2. The method of claim 1 , wherein processing the plurality of multimodal signatures comprises executing a design information extraction operation.

3. The method of claim 1 , wherein processing the plurality of multimodal signatures comprises executing a design information recovery operation.

4. The method of claim 1 , wherein processing the plurality of multimodal signatures comprises executing a spatial risk scoring operation.

5. The method of claim 1 , wherein the data representations include hyperspectral-multimodal scans of the object of interest, assessments of a bill of materials of the object of interest, determinations of how components are connected within the object of interest, or vulnerability information for the object of interest.

6. The method of claim 1 , wherein processing the plurality of multimodal signatures comprises labeling a first subset of an unlabeled dataset of the plurality of multimodal signatures and training an artificial neural network on the labeled first subset.

7. The method of claim 1 , wherein processing the plurality of multimodal signatures comprises:
generating a plurality of labeled signatures from the plurality of multimodal signatures;
grouping each of the plurality of labeled signatures into training tiles of a fixed physical size; and
training an artificial neural network to identify components of the object of interest based on the training tiles.

8. The method of claim 1 , wherein processing the plurality of multimodal signatures comprises:
selecting training data having m modalities from the plurality of multimodal signatures;
grouping the training data into training tiles of a fixed physical size; and
training m conditional generative adversarial networks to generate candidate tiles for each of the m modalities.

9. A system comprising:
a memory configured to store a software for a multimodal evaluation engine; and
one or more processors coupled to the memory, the one or more processors configured to execute the software for the multimodal evaluation engine to cause the system to perform:
receiving a plurality of multimodal signatures of an object of interest from one or more inspection elements, wherein the plurality of multimodal signatures comprise visible light, infrared, electromagnetic interference, and laser profilometry;
processing the plurality of multimodal signatures to transform the plurality of multimodal signatures into one or more formats;
generating data representations of the one or more formats; and
detecting whether one or more anomalies are present within the object of interest based on the data representations.

10. The system of claim 9 , wherein processing the plurality of multimodal signatures comprises executing a design information extraction operation.

11. The system of claim 9 , wherein processing the plurality of multimodal signatures comprises executing a design information recovery operation.

12. The system of claim 9 , wherein processing the plurality of multimodal signatures comprises executing a spatial risk scoring operation.

13. The system of claim 9 , wherein the data representations include hyperspectral-multimodal scans of the object of interest, assessments of a bill of materials of the object of interest, determinations of how components are connected within the object of interest, or vulnerability information for the object of interest.

14. The system of claim 9 , wherein processing the plurality of multimodal signatures comprises labeling a first subset of an unlabeled dataset of the plurality of multimodal signatures and training an artificial neural network on the labeled first subset.

15. The system of claim 9 , wherein processing the plurality of multimodal signatures comprises:
generating a plurality of labeled signatures from the plurality of multimodal signatures;
breaking/grouping each of the plurality of labeled signatures into training tiles of a fixed physical size; and
training an artificial neural network to identify components of the object of interest based on the training tiles.

16. The system of claim 9 , wherein processing the plurality of multimodal signatures comprises:
selecting training data having m modalities from the plurality of multimodal signatures;
breaking/grouping the training data into training tiles of a fixed physical size; and
training m conditional generative adversarial networks to generate candidate tiles for each of the m modalities.

17. A computer readable medium storing a software for a multimodal evaluation engine, the software being executable by one or more processors to cause the multimodal evaluation engine to perform:
receiving a plurality of multimodal signatures of an object of interest from one or more inspection elements, wherein the plurality of multimodal signatures comprise visible light, infrared, electromagnetic interference, and laser profilometry;
processing the plurality of multimodal signatures to transform the plurality of multimodal signatures into one or more formats;
generating data representations of the one or more formats; and
detecting whether one or more anomalies are present within the object of interest based on the data representations.

Assignees

• Battelle Memorial Institute

Inventors

• George Anthony
• Darby Nicholas
• Bellay Jeremy
• Collins David
• Liszewski Katie
• Rahimi Amir

Classifications

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/0475

  Generative networks · CPC title

• G06N3/0895

  Weakly supervised learning, e.g. semi-supervised or self-supervised learning · CPC title

• G06N3/09

  Supervised learning · CPC title