Deep neural network training method and system, electronic device

Abstract

Embodiments of the invention disclose a deep neural network training method and system and electronic equipment. The method comprises the following steps of: in a forward propagation process, carryingout scene analysis detection on a sample image by utilizing a deep neural network model so as to obtain a first scene analysis prediction result output by a middle network layer and a second scene analysis prediction result output by a tail network layer; determining a first difference between the first scene analysis prediction result and scene analysis labeling information of the sample image and a second difference between the second scene analysis prediction result and the scene analysis labeling information of the sample image; and in a counter-propagation process, adjusting parameters of a first network layer according to the first difference and adjusting parameters of the a second network layer according to the first difference and the second difference, wherein the first networklayer comprises at least one network layer between the middle network layer and the tail network layer, and the second network layer comprises other network layers except the first network layer. According to the method and system and the electronic equipment, better network model optimization results can be obtained.

Description

technical field [0001] The invention relates to computer vision technology, in particular to a deep neural network training method, shampoo and electronic equipment. Background technique [0002] For neural networks, it can be clearly found that the expressiveness and performance of the network increases with the increase of network depth. However, the deeper the network, the better. In addition to the problem of computational cost, when the network depth is deep, continuing to increase the number of network layers will not improve network performance, but may degrade it. In addition, for deep neural networks, due to reasons such as gradient disappearance, how to train a deep neural network has always been a problem that plagues people. Contents of the invention [0003] An embodiment of the present invention provides a technical solution for training a deep neural network. [0004] According to an aspect of an embodiment of the present invention, a neural network train...

AI Technical Summary

Problems solved by technology

In addition to the calculation cost problem, when the network depth is deep, continuing to increase the number of network layers will not improve network performance, but may degrade

In addition, for deep neural networks, due to reasons such as gradient disappearance, how to train a deep neural network has always been a problem that plagues people.

Images