Neural network model training method, electronic device and storage medium

Abstract

The invention discloses a neural network model training method. The method comprises a step of obtaining a first image feature and a second image feature, wherein the first image feature is an image feature outputted by a picture A through a trained model, and the second image feature is an image feature outputted by the image A through a model to be trained, a step of obtaining a classification probability which is a classification probability outputted by the second image feature through a classification layer of the model to be trained, and a step of updating the parameters of the model tobe trained until the second image feature and the first image feature are fitted according to the first image feature, the second image feature and the classification probability. The invention also discloses an electronic device and a storage medium. The invention provides the neural network model training method, the electronic device and the storage medium, the parameters of the model to be trained are updated according to the first image feature, the second image feature and the classification probability, a small model with the same precision of a trained large model is obtained through training, and the calculation speed and accuracy are ensured.

Description

technical field [0001] The invention relates to the field of model training, in particular to a training method of a neural network model, electronic equipment and a storage medium. Background technique [0002] The convolutional neural network includes multiple convolutional layers, activation function layers, and downsampling layers. Its local sampling and weight sharing characteristics make it possible to obtain translation-invariant features. The extracted features far exceed the traditional hand-designed features, which greatly improves the The accuracy of machine vision tasks such as recognition, detection, and segmentation has fewer parameters than traditional neural networks and is easy to train. [0003] With the development of technology, the trend of convolutional neural network is that the deeper the model, the higher the accuracy. The improvement of accuracy makes many visual tasks practical, but also brings a huge amount of calculation. The deeper the model, th...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the deficiencies of the prior art, one of the purposes of the present invention is to provide a training method for a neural network model to solve the problem that the existing model compression technology affects the calculation speed and calculation accuracy

[0006] The second object of the present invention is to provide an electronic device to solve the problem that the existing model compression technology affects the calculation speed and calculation accuracy

Images