Method for reducing transmission consumption of neural network model update on mobile device

Abstract

The invention discloses a method for reducing transmission consumption of neural network model updates at a mobile equipment side. The method comprises the following steps: step 1, mobile equipment selects predicted picture information to upload to the cloud for relearning, and thus the consumption cost of uploading data can be reduced, and the performance improvement brought by relearning cannot be influenced; step 2, the new data uploaded by the mobile equipment is mixed with an old data set to form a larger training set at the cloud, a neural network model is retrained to achieve the learning of the new data, and the problem of catastrophic forgetting in incremental learning of neural networks can be avoided; and step 3, a small amount of weight in a new model is extracted at the cloud and is transmitted to the mobile equipment, an old model that has been deployed on the mobile equipment is updated to ensure that the old model on the mobile equipment can reach the identification performance of the new model, and the cost of data transmission generated when updating old neural network models in the mobile equipment can be reduced. The method disclosed by the invention has the technical effect of effectively reducing the transmission consumption cost of the neural network model updates at the mobile equipment side.

Description

technical field [0001] The invention belongs to the technical field of mobile equipment sharing, and in particular relates to a method for reducing the transmission consumption of neural network model update at the mobile equipment end. Background technique [0002] Neural networks include convolutional neural networks, recurrent networks, etc. At present, neural networks have been applied to different large-scale machine learning problems such as speech recognition, image recognition, and natural speech processing. Internet has become an inevitable trend. However, due to the large amount of calculation and large memory consumption of the neural network, the use and training of the neural network on the mobile terminal cannot be undertaken. The task is sent to the cloud for execution, and the result is returned to the mobile terminal. However, this method requires the device to upload data to the cloud and download results from the cloud in real time, so there is network ...

AI Technical Summary

Problems solved by technology

[0004] Although the mobile terminal can execute the neural network, the mobile terminal still needs a powerful server in the cloud to complete the training of the neural network, so the neural network model transplanted to the mobile terminal is solidified and loses the ability to re-learn

Images