Training an auto-encoder on a single class

• US 10,607,135 B2
• Filed: 12/27/2017
• Issued: 03/31/2020
• Est. Priority Date: 10/19/2017
• Status: Active Grant

• Current Assignee: General Electric Company
• Original Assignee: General Electric Company
• Inventors: Zhang, Min, Avinash, Gopal Biligeri
• Primary Examiner(s): Misleh, Justin P.
• Application Number:
  US15/854,980
• Publication Number:
  US 20190122075A1
• Time in Patent Office: 825 Days
• Field of Search: None
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06N 20/20 Ensemble learning
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06T 2207/10072 Tomographic images
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/20084 Artificial neural networks [ANN]
• CPC Class Codes: G06T 2207/30016 Brain
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 10/772 Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries
• CPC Class Codes: G06V 2201/03 Recognition of patterns in medical or anatomical images
• CPC Class Codes: G16H 30/40 for processing medical images, e.g. editing
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: G16H 50/50 for simulation or modelling of medical disorders

Abstract

Systems and techniques for training an auto-encoder on a single class are presented. In one example, a system trains an auto-encoder based on first data associated with a first class to generate a trained auto-encoder. The system also applies, using a multiplier, gain data indicative of a gain value to second data associated with the first class and third data associated with a second class to generate enhanced input data that represents a differentiation between the second data associated with the first class and the third data associated with the second class. An input enhancer comprises the trained auto-encoder and the multiplier. Furthermore, the system trains a convolutional neural network based on the enhanced input data to generate a trained convolutional neural network. The system also classifies the first class and the second class based on the input enhancer and the trained convolutional neural network.

Claims

1. A machine learning system, comprising:

a memory that stores computer executable components;

a processor that executes the computer executable components stored in the memory, wherein the computer executable components comprise;

an auto-encoder component that trains an auto-encoder based on first data associated with a first class to generate a trained auto-encoder;

a multiplier component that applies, using a multiplier, gain data indicative of a gain value to second data associated with the first class and third data associated with a second class to generate enhanced input data that represents a differentiation between the second data associated with the first class and the third data associated with the second class, wherein an input enhancer comprises the trained auto-encoder and the multiplier;

a convolutional neural network component that trains a convolutional neural network based on the enhanced input data to generate a trained convolutional neural network; and

a classification component that classifies the first class and the second class based on the input enhancer and the trained convolutional neural network.

2. The machine learning system of claim 1, wherein the first data is a first set of images associated with the first class, and wherein the auto-encoder component trains the auto-encoder based on the first set of images associated with the first class.

3. The machine learning system of claim 2, wherein the second data is a second set of images associated with the first class and the third data is a third set of images associated with the second class, and wherein the multiplier component that applies the gain data to the second set of images and the third set of images to generate the enhanced input data.

4. The machine learning system of claim 2, wherein the classification component classifies the first class and the second class from a new set of images based on the input enhancer and the trained convolutional neural network.

5. The machine learning system of claim 1, wherein the auto-encoder component compares reconstructed second data and reconstructed third data to the second data and the third data to determine loss function data indicative of a loss function for the auto-encoder, and wherein the auto-encoder component generates the trained auto-encoder based on the loss function data.

6. The machine learning system of claim 1, wherein the convolutional neural network performs a plurality of sequential and/or parallel downsampling and upsampling of the enhanced input data associated with convolutional layers of the convolutional neural network.

7. The machine learning system of claim 1, wherein the classification component determines a classification for new data associated with the first class and the second class based on the input enhancer and the trained convolutional neural network.

8. The machine learning system of claim 7, wherein the computer executable components further comprise:

a visualization component that generates a multi-dimensional visualization associated with the classification for the new data.

9. A method, comprising using a processor operatively coupled to memory to execute computer executable components to perform the following acts:

inputting first image data associated with a first class;

training an auto-encoder based on the first class associated with the first image data to generate a trained auto-encoder;

inputting second image data associated with the first class and third image data associated with a second class to the trained auto-encoder;

applying, using a multiplier, gain data indicative of a gain value to the second image data and the third image data to generate enhanced input data that represents a differentiation between the second image data associated with the first class and the third image data associated with the second class, wherein an input enhancer comprises the trained auto-encoder and the multiplier;

training a convolutional neural network based on the enhanced input data to generate a trained convolutional neural network; and

identifying the first class and the second class from new image data based on the input enhancer and the trained convolutional neural network.

10. The method of claim 9, wherein the training the auto-encoder comprises training the auto-encoder based on the first image data associated with the first class to generate reconstructed first image data.

11. The method of claim 10, wherein the training the auto-encoder comprises comparing the reconstructed first image data to the first image data to determine loss function data indicative of a loss function for the auto-encoder.

12. The method of claim 11, wherein the training the auto-encoder comprises generating the trained auto-encoder based on the loss function data.

13. The method of claim 9, wherein the applying the gain data comprises applying the gain data to reconstructed second image data and reconstructed third image data to generate the enhanced input data.

14. The method of claim 9, wherein the identifying comprises identifying one or more outlier images that satisfy a defined criterion from the first class based on the input enhancer and the trained convolutional neural network.

15. The method of claim 9, wherein the training the convolutional neural network comprises performing a plurality of sequential and/or parallel downsampling and upsampling of the enhanced input data associated with convolutional layers of the convolutional neural network.

16. The method of claim 9, further comprising:

determining a classification for the first class and the second class from new data based on the input enhancer and the trained convolutional neural network.

17. The method of claim 16, further comprising:

generating a multi-dimensional visualization associated with the classification for the new data.

18. A computer readable storage device comprising instructions that, in response to execution, cause a system comprising a processor to perform operations, comprising:

training an auto-encoder based on first image data associated with a first class to generate a trained auto-encoder;

applying, using a multiplier, gain data indicative of a gain value to second image data associated with the first class and third image data associated with a second class to generate enhanced input data that represents a differentiation between the second image data associated with the first class and the third image data associated with the second class, wherein an input enhancer comprises the trained auto-encoder and the multiplier;

training a convolutional neural network based on the enhanced input data to generate a trained convolutional neural network; and

identifying the first class and the second class from new image data based on the input enhancer and the trained convolutional neural network.

19. The computer readable storage device of claim 18, wherein the identifying comprises identifying one or more outlier images from the first class based on the input enhancer and the trained convolutional neural network.

20. The computer readable storage device of claim 18, wherein the training the auto-encoder comprises comparing reconstructed first image data to the first image data to minimize loss function data indicative of a loss function for the auto-encoder.