Edge-based adaptive machine learning for object recognition

Patent metadata

Field	Value
Publication number	US-11017271-B2
Application number	US-201715801791-A
Country	US
Kind code	B2
Filing date	Nov 2, 2017
Priority date	Oct 24, 2016
Publication date	May 25, 2021
Grant date	May 25, 2021

Abstract

Examples of techniques for interactive generation of labeled data and training instances are provided. According to one or more embodiments of the present invention, a computer-implemented method for interactive generation of labeled data and training instances includes presenting, by the processing device, control labeling options to a user. The method further includes selecting, by a user, one or more of the presented control labeling options. The method further includes selecting, by a processing device, a representative set of unlabeled data samples based at least in part on the control labeling options selected by the user. The method further includes generating, by a processing device, a set of suggested labels for each of the unlabeled data samples.

First claim

What is claimed is:

1. A computer-implemented method for interactive generation of labeled data and training instances, the method comprising:
generating, by a processing device, a plurality of bounding boxes for an image stored on a first user's mobile device;
removing, by the processing device, any bounding box of the plurality of bounding boxes that has an area greater than forty-five percent of the image, and removing any bounding box of the plurality of bounding boxes that has an area less than one percent of the digital image;
presenting, by the processing device, control labeling options for objects in the plurality of bounding boxes;
receiving, by the processing device, one or more selected control labeling options of the presented control labeling options, wherein each of the one or more selected control labeling options describes a respective object in a bounding box of the plurality of bounding boxes;
generating, by the processing device, a training instance comprising a label of the one or more selected control labeling options and a machine learning extracted feature of the object associated with the label;
transmitting, to a second user's mobile computing device, the feature and associated label;
training an adapted model using the feature and the associated label at the second user's mobile computing device; and
sharing the adapted model with the first user's mobile computing device.

2. The computer-implemented method of claim 1 , wherein the set of suggested labels for each of the unlabeled images are generated from a generic machine learning model.

3. The computer-implemented method of claim 1 further comprising presenting the generated set of suggested labels to a third user.

4. The computer-implemented method of claim 3 further comprising receiving a selected label for each unlabeled image from the third user.

5. The computer-implemented method of claim 4 further comprising verifying the selected label for truthfulness.

6. The computer-implemented method of claim 4 further comprising extracting the selected label and at least one associated feature from the unlabeled image.

7. The computer-implemented method of claim 6 further comprising storing the selected label and the at least one associated feature as a training instance.

8. The computer-implemented method of claim 1 , wherein the representative set of unlabeled images is selected from one or more devices by an algorithm.

9. The computer-implemented method of claim 8 , wherein the algorithm is selected from the group consisting of a distributed clustering algorithm and a principle component analysis algorithm.

10. The computer-implemented method of claim 1 , wherein the labeling options are selected from the group consisting of an amount of time spent labeling, a number of samples to be labeled, and contextual parameters.

11. The computer-implemented method of claim 1 , further comprising:
dividing the digital image into regions; and
generating a threshold number of bounding boxes for each respective region.

Assignees

• IBM

Inventors

• Desai Nirmit V
• Li Dawei
• Salonidis Theodoros

Classifications

• G06V30/19167

  Active pattern learning · CPC title

• G06V30/19173

  Classification techniques · CPC title

• G06V10/82

  using neural networks · CPC title

• G06F18/2178

  based on feedback of a supervisor · CPC title

• G06N20/00 Primary

  Machine learning · CPC title