Multilayer perceptron artificial nerve network based on residual error network

Abstract

The invention discloses a multilayer perceptron artificial nerve network based on a residual error network; the multilayer perceptron artificial nerve network based on the residual error network comprises a plurality of network module structures; a full connection mode is employed to replace the convolution in the residual error nerve network; Neuron units in the network module structures can obtain the output of the complete residual error module through the output of each hidden layer, wherein the output of each hidden layer refers to si=ReLU[BN(neti)]; the complete residual error module output refers to oi=ReLU[BN(neti+1)+neti]. The multilayer perceptron artificial nerve network is based on the residual error network, small in computational complexity, high in accuracy, and can be applied to more fields besides the image field.

Description

technical field [0001] The invention relates to the field of bionic network computing, in particular to a residual network-based multi-layer perceptron artificial neural network. Background technique [0002] Artificial Neural Network (Artificial Neural Network) is a bionic network. By imitating the central nervous system of the biological brain, a mathematical model or computational model with function estimation and analysis is established. It is usually used in the fields of machine learning and cognitive learning. Multilayer Perception (Multilayer Perception), also known as the forward propagation network, was first proposed by Paul J.Werbos in his doctoral thesis in 1974. It is a typical deep learning structure, including an input layer and an output layer. and hidden layers. The number and complexity of the hidden layer determine the ability of the network, and an overly complex network is prone to overfitting. How to correctly design the hidden layer is the difficult...

AI Technical Summary

Problems solved by technology

However, in the process of use, the residual network is often used in conjunction with the convolutional neural network (Convolutional Neural Network). The chain derivation in the convolution operation and the backpropagation process will cause a large amount of calculations in the training process; at the same time, the convolution Its own characteristics make it more suitable for the operation of images, but for other fields such as speech signals and natural language processing, the effect is slightly weaker

Images