Unlock instant, AI-driven research and patent intelligence for your innovation.

Multi-task learning method and device

A multi-task learning and task technology, applied in neural learning methods, neural architectures, biological neural network models, etc., can solve problems such as wireless communication and multi-task learning implementation solutions that are not given

Pending Publication Date: 2021-12-24
HUAWEI TECH CO LTD
View PDF0 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, at present, multi-task learning has not been applied to wireless communication, and no implementation scheme for combining wireless communication with multi-task learning has been given.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Multi-task learning method and device
  • Multi-task learning method and device
  • Multi-task learning method and device

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0126] Embodiment one, see Figure 8 The neural network architecture shown, lossless compression and joint training of 3 encoders.

[0127] The input of the neural network: source X and source Y.

[0128] Training constraints: R 0 +R 1 +R 2 =H(X,Y),R 0 is the coding rate for source X and source Y, R 1 is the coding rate of the source X, R 2 is the coding rate of source Y, H(X, Y) is the entropy of source X and source Y, and this constraint can be regarded as a newly designed objective function.

[0129] The goal of training: Losslessly reconstruct source X and source Y.

[0130] The training process includes the following steps:

[0131] Step 1.1: Divide the data of source X and source Y into two parts, one part of data is used as training data, and the other part of data is used as test data.

[0132] For example, the information source X may be the information source of task 1, and the information source Y may be the information source of task 2.

[0133] The train...

Embodiment 2

[0141]Embodiment 2, lossless compression and three encoders are trained separately.

[0142] The input of the neural network: source X and source Y.

[0143] Training constraints, according to different needs, different constraints can be set during the separate training of the encoder:

Embodiment 21

[0144] Example 2.1, R 0 =H(X,Y),R 1 = R 2 = 0, at this time, encoder 0 is trained, and encoder 1 and encoder 2 (that is, private encoders) are not trained.

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The embodiment of the invention relates to a multi-task learning method and device. The method comprises the steps of inputting training data in information sources of multiple tasks into a coding network to be coded, and obtaining coded training data, wherein the coded training data comprise feature information of the multiple tasks, and the information sources of the multiple tasks comprise labeling results of the training data; inputting the coded training data into a decoding network for decoding to obtain a prediction result of the training data; according to the feature information of the plurality of tasks and a constraint condition, adjusting a first neural network parameter of the coding network, the constraint condition being a condition satisfied when the coding network reaches a convergence state; and adjusting a second neural network parameter of the decoding network according to an error between an annotation result of the training data and a prediction result of the training data so as to realize application of multi-task learning in wireless communication.

Description

technical field [0001] The present application relates to the technical field of neural networks, in particular to a multi-task learning method and device. Background technique [0002] Existing neural networks generally learn a single task, which has certain limitations. For example, a neural network trained for a certain task cannot be applied to another task, and its generalization ability is poor. [0003] Therefore, the multi-task learning method was born. Multi-task learning is a learning method that uses the correlation between multiple tasks to promote each other and learn multiple tasks in parallel. With the development of deep learning, the combination of multi-task learning and deep neural network has been successfully applied to natural language processing, computer vision and other fields. However, multi-task learning has not been applied to wireless communication at present, and no implementation scheme for combining wireless communication with multi-task lear...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
IPC IPC(8): G06N3/04G06N3/08
CPCG06N3/08G06N3/045G06N3/096G06N3/084G06N3/0455
Inventor 孔垂丽张公正李榕
Owner HUAWEI TECH CO LTD