Edge-based adaptive machine learning for object recognition

Patent metadata

Field	Value
Publication number	US-11288551-B2
Application number	US-201715727764-A
Country	US
Kind code	B2
Filing date	Oct 9, 2017
Priority date	Oct 24, 2016
Publication date	Mar 29, 2022
Grant date	Mar 29, 2022

Abstract

Examples of techniques for adaptive model training are provided. According to one or more embodiments of the present invention, a computer-implemented method for adaptive model training includes generating, by a processing system, a training instance based at least in part on a plurality of images that match a contextual specification of a target visual domain. The method further includes extracting, by the processing system, objects from one of the plurality of images. The method further includes for each extracted object, generating, by the processing system, a plurality of machine learning model features and label recommendations for a user.

First claim

What is claimed:

1. A system for adaptive neural network model training, the system comprising:
a memory comprising computer readable instructions; and
a processing system for executing the computer readable instructions for performing a method, the method comprising:
receiving, by the processing system, a set of attributes of a target visual domain for a local image stored on a first user device, the local image having an image area that includes a plurality of visual objects;
determining, by the processing system, adaption model criteria based on the set of attributes of the target visual domain;
determining a number of initial object proposals for the visual objects based on the set of attributes of the target visual domain;
constraining the initial object proposals based on mobile contextual information to determine a reduced number of object proposals, the mobile contextual information including one or combination of a location of the first user device, historical contextual information corresponding to previously detected objects of interest at the location of the first user device, and a number of label recommendations describing the visual object for selection, the number of label recommendations requested by a user of the first user device;
determining a proposed bounding box for each of the reduced object proposals and determining a bounding area for each proposed bounding box;
determining a first percentage area of the image area and a second percentage area of the image area, and removing targeted proposed bounding boxes having a bounding area that is less than the first percentage area or greater than the second percentage area;
extracting, by the processing system, visual objects corresponding to the remaining bounding boxes generated for the reduced object proposals of the local image stored on the first user device;
for each extracted visual object, generating, by the processing system, a plurality of machine learning model features and the number of label recommendations requested by the user,
creating a training set based on the plurality of machine learning model features and a received selected labels from the first user device; and
training a new adaptive model using the training set.

2. The system of claim 1 , wherein the training set is stored in an adaptation database.

3. The system of claim 2 , wherein the new adapted model is stored in the adaptation database.

4. The system of claim 1 , wherein:
the set of attributes includes one or more of location information, time information, and weather information.

5. The system of claim 1 , wherein generating the plurality of machine learning model features and label recommendations further comprises inputting the extracted object images into a generic machine learning model.

6. A computer program product for adaptive neural network model training, the computer program product comprising:
a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processing system to cause the processing system to perform a method comprising:
receiving a set of attributes of a target visual domain for a local image stored on a first user device, the local image having an image area that includes a plurality of visual objects;
determining adaption model criteria based on the set of attributes of the target visual domain;

determining a number of initial object proposals for the visual objects based on the set of attributes of the target visual domain;
constraining the initial object proposals based on mobile contextual information to determine a reduced number of object proposals, the mobile contextual information including one or combination of a location of the first user device, historical contextual information corresponding to previously detected objects of interest at the location of the first user device, and a number of label recommendations describing the visual object for selection, the number of label recommendations requested by a user of the first user device;
determining a proposed bounding box for each of the reduced object proposals and determining a bounding area for each proposed bounding box;
determining a first percentage area of the image area and a second percentage area of the image area, and removing targeted proposed bounding boxes having a bounding area that is less than the first percentage area or greater than the second percentage area;
extracting visual objects from the bounding boxes generated for the local image stored on the first user device;
for each extracted visual object, generating, by the processing system, a plurality of machine learning model features and label recommendations describing the visual objects for selection by a user of the first user device;
creating a training set based on the plurality of machine learning model features and a received selected labels from the first user device; and
training a new adaptive model using the training set.

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