Variational gradient flow

Patent metadata

Field	Value
Publication number	US-11475304-B2
Application number	US-202016872907-A
Country	US
Kind code	B2
Filing date	May 12, 2020
Priority date	May 12, 2020
Publication date	Oct 18, 2022
Grant date	Oct 18, 2022

Abstract

According to embodiments of the present disclosure, methods of and computer program products for operating a plurality of classifiers are provided. A plurality of input entities are read, each input entity having an associated target label. The input entities are provided to a first classifier, and a category of each input entity is obtained therefrom. A feature map is determined for each input entity. Each feature map is provided to each of a set of classifiers, and an assigned label is obtained for each feature map from each of the set of classifiers. Each classifier is associated with one of the categories. For each classifier, the assigned label for each feature map is compared to the target labels to determine a plurality of gradients. The plurality of gradients are masked according to each category, yielding a masked set of gradients for each category. Each classifier is trained according its associated masked gradients.

First claim

What is claimed is:

1. A method comprising:
reading a plurality of input entities, each of the input entities having an associated target label;
providing the input entities to a first classifier, and obtaining therefrom a category of each of the input entities;
determining a feature map for each of the input entities;
providing each of the feature maps to each of a set of classifiers, and obtaining from each of the set of classifiers an assigned label for each of the plurality of feature maps, each of the set of classifiers being associated with one of the categories;
comparing, for each of the set of classifiers, the assigned label for each of the plurality of feature maps to the target labels, to determine a plurality of gradients;
masking the plurality of gradients according to each category, yielding a masked set of gradients for each of the categories; and
training each of the set of classifiers according its associated masked gradients.

2. The method of claim 1 , further comprising:
providing an input entity to the first classifier, and obtaining therefrom a category of the input entity;
determining a feature map for the input entity;
providing the feature map to each of a set of classifiers, and obtaining from each of the set of classifiers a label for the feature map;
masking the labels according to the category; and
outputting the masked labels.

3. The method of claim 1 , wherein each of the plurality of input entities comprises an image.

4. The method of claim 1 , wherein the first classifier is pre-trained.

5. The method of claim 1 , wherein the first classifier comprises an artificial neural network.

6. The method of claim 1 , wherein each of the set of classifiers comprises an artificial neural network.

7. The method of claim 1 , wherein each of the target labels is a member of one of the categories.

8. A computer program product for operating a plurality of classifiers, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to perform a method comprising:
reading a plurality of input entities, each of the input entities having an associated target label;
providing the input entities to a first classifier, and obtaining therefrom a category of each of the input entities;
determining a feature map for each of the input entities;
providing each of the feature maps to each of a set of classifiers, and obtaining from each of the set of classifiers an assigned label for each of the plurality of feature maps, each of the set of classifiers being associated with one of the categories;
comparing, for each of the set of classifiers, the assigned label for each of the plurality of feature maps to the target labels, to determine a plurality of gradients;
masking the plurality of gradients according to each category, yielding a masked set of gradients for each of the categories; and
training each of the set of classifiers according its associated masked gradients.

9. The computer program product of claim 8 , the method further comprising:
providing an input entity to the first classifier, and obtaining therefrom a category of the input entity;
determining a feature map for the input entity;
providing the feature map to each of a set of classifiers, and obtaining from each of the set of classifiers a label for the feature map;
masking the labels according to the category; and
outputting the marks labels.

10. The computer program product of claim 8 , wherein each of the plurality of input entities comprises an image.

11. The computer program product of claim 8 , wherein the first classifier is pre-trained.

12. The computer program product of claim 8 , wherein the first classifier comprises an artificial neural network.

13. The computer program product of claim 8 , wherein each of the set of classifiers comprises an artificial neural network.

14. The computer program product of claim 8 , wherein each of the target labels is a member of one of the categories.

15. A method comprising:
inputting data into both: i) a plurality of fine-grained classifiers and ii) a course-grained classifier, the coarse grained classifier configured to categorize the input data among a plurality of masking modules, each of the plurality of masking modules corresponding to one of the plurality of fine-grained classifiers;
during a training phase, computing a loss function corresponding to each pair of the fine-grained classifiers and masking modules; and
during an inference phase, assigning the data to a particular class in view of output from the plurality of fine-grained classifiers and the plurality of masking modules.

16. The method of claim 15 , wherein the data comprise a plurality of images.

17. The method of claim 15 , wherein the coarse classifier is pre-trained.

18. The method of claim 15 , wherein the coarse classifier comprises an artificial neural network.

19. The method of claim 15 , wherein each of the plurality of fine-grained classifiers comprises an artificial neural network.

Assignees

• IBM

Inventors

• Kornuta Tomasz
• Ozcan Ahmet Serkan
• Rajan Deepta
• Asseman Alexis
• Shivade Chaitanya

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/084 Primary

  Backpropagation, e.g. using gradient descent · CPC title

• G16H30/40

  for processing medical images, e.g. editing · CPC title

• G16H50/20

  for computer-aided diagnosis, e.g. based on medical expert systems · CPC title

• G06N3/08 Primary

  Learning methods · CPC title