Unsupervised training sets for content classification

Patent metadata

Field	Value
Publication number	US-10360498-B2
Application number	US-201414575547-A
Country	US
Kind code	B2
Filing date	Dec 18, 2014
Priority date	Dec 18, 2014
Publication date	Jul 23, 2019
Grant date	Jul 23, 2019

Abstract

Various embodiments of the present disclosure include systems, methods, and non-transitory computer storage media configured to identify a set of training content items, each of the set of training content items comprising video content. A category may be assigned to each of the set of training content items. A plurality of variations may be provided to the each of the set of training content items. A first content recognition module may be trained in an unsupervised process to associate the plurality of variations of the each of the set of training content items with the category assigned to the each of the set of training content items. A classification layer may be generated based on the training the first content recognition module in the unsupervised process. A second content recognition module may be trained in a supervised process based on the classification layer.

First claim

What is claimed:

1. A computer implemented method comprising:
identifying, by a computing system, a set of training content items, each of the set of training content items comprising video content;
assigning, by the computing system, a category to each of the set of training content items;
providing, by the computing system, a plurality of variations of the each of the set of training content items;
training, by the computing system, a first instance of a content recognition module comprising a first convolutional neural network in an unsupervised process to associate the plurality of variations of the each of the set of training content items with the category assigned to the each of the set of training content items;
generating, by the computing system, a classification layer of the first instance of the content recognition module from the training the first instance of the content recognition module in the unsupervised process, wherein the classification layer is trained to recognize invariances in the each of the set of training content items and the plurality of variations of the each of the set of training content items;
replacing, by the computing system, a classification layer of a second instance of the content recognition module comprising a second convolutional neural network with the classification layer of the first instance of the content recognition module to provide the second instance of the content recognition module with a new classification layer; and
training, by the computing system, the second instance of the content recognition module in a supervised process based on the new classification layer, wherein the training the second instance of the content recognition module includes updating one or more layers of the second convolutional neural network by performing a backpropagation based on the new classification layer.

2. The method of claim 1 , wherein the plurality of variations comprises a variation of an object in the video content.

3. The method of claim 1 , wherein the training the second instance of the content recognition module in the supervised process comprises associating the each of the set of training content items with a semantic sequence corresponding to the category assigned to the each of the set of training content items.

4. The method of claim 1 , wherein the plurality of variations comprises at least one geometric variation of the each of the set of training content items.

5. The method of claim 1 , wherein the plurality of variations comprises at least one of a rotation, a translation, a rescaling, a color change, a geometric modification, or a filtering of the each of the set of training content items.

6. The method of claim 1 , wherein the plurality of variations comprises a variation of a perspective, lighting, or motion of an object.

7. The method of claim 1 , further comprising using the second instance of the content recognition module to classify a set of evaluation content items.

8. The method of claim 7 , wherein the set of evaluation content items includes content items uploaded by users of a social networking system.

9. A system comprising:
at least one processor;
a memory storing instructions configured to instruct the at least one processor to perform:
identifying, by a computing system, a set of training content items, each of the set of training content items comprising video content;
assigning, by the computing system, a category to each of the set of training content items;
providing, by the computing system, a plurality of variations of the each of the set of training content items;
training, by the computing system, a first instance of a content recognition module comprising a first convolutional neural network in an unsupervised process to associate the plurality of variations of the each of the set of training content items with the category assigned to the each of the set of training content items;
generating, by the computing system, a classification layer of the first instance of the content recognition module from the training the first instance of the content recognition module in the unsupervised process, wherein the classification layer is trained to recognize invariances in the each of the set of training content items and the plurality of variations of the each of the set of training content items;
replacing, by the computing system, a classification layer of a second instance of the content recognition module comprising a second convolutional neural network with the classification layer of the first instance of the content recognition module to provide the second instance of the content recognition module with a new classification layer; and
training, by the computing system, the second instance of the content recognition module in a supervised process based on the new classification layer, wherein the training the instance of the content recognition module includes updating one or more layers of the second convolutional neural network by performing a backpropagation based on the new classification layer.

10. The system of claim 9 , wherein the plurality of variations comprises a variation of an object in the video content.

11. The system of claim 9 , wherein the training the second convolutional neural network in the supervised process comprises associating each of the set of training content items with a semantic sequence corresponding to the category assigned to the each of the set of training content items.

12. The system of claim 9 , wherein the plurality of variations comprises a variation of a perspective, lighting, or motion of an object.

13. The system of claim 9 , wherein the instructions are configured to instruct the at least one processor to further perform:
using the second instance of the content recognition module to classify a set of evaluation content items.

14. The system of claim 13 , wherein the set of evaluation content items includes content items uploaded by users of a social networking system.

15. A non-transitory computer storage medium storing computer-executable instructions that, when executed, cause a computer system to perform a computer-implemented method comprising:
identifying, by a computing system, a set of training content items, each of the set of training content items comprising video content;
assigning, by the computing system, a category to each of the set of training content items;
providing, by the computing system, a plurality of variations of the each of the set of training content items;
training, by the computing system, a first instance of a content recognition module comprising a first convolutional neural network in an unsupervised process to associate the plurality of variations of the each of the set of training content items with the category assigned to the each of the set of training content items;
generating, by the computing system, a classification layer of the first instance of the content recognition module from the training the first instance of the content recognition module in the unsupervised process, wherein the classification layer is trained to recognize invariances in the each of the set of training content items and the plurality of variations of the each of the set of training content items;
replacing, by the computing system, a classification layer of a second instance of the content recognition module comprising a second convolutional neural network with the classification layer of the first instance of the content recognition module to provide the second instance of the content recognition module with a new classification layer; and
training, by the c

Assignees

• Facebook Inc

Inventors

• Fergus Robert D
• Bourdev Lubomir
• Paluri Balamanohar
• Sukhbaatar Sainbayar

Classifications

• G06N3/08 Primary

  Learning methods · CPC title

• G06N20/00

  Machine learning · CPC title

• G06N3/088 Primary

  Non-supervised learning, e.g. competitive learning · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title