Improvement And Application Of Autoencoders

With the rapid development of artificial intelligence,we have also improved our understanding to the world.As an important part of artificial intelligence,machine learning can model the word which has an important function.In recent years,deep learning has received much attention from the academic community.As one of the deep learning algorithms,deep autoencoder is widely used in daily life.While studying the deep autoencoder algorithm and the classifier algorithm,this paper also explores the way to improve the deep autoencoder and the combination of deep autoencoder and other algorithms,and compares the differences of various algorithms through experiments.The main research contents of this paper are as follows:1.First,this paper introduces typical autoencoder model,sparse autoencoder,denoising autoencoder and stack autoencoder.2.And then,the advantages and disadvantages of various classifiers are compared and analyzed,including support vector machine algorithm,random forest algorithm,gradient boosting decison tree algorithm,decision tree algorithm,K-nearest neighbor algorithm.For the small sample dataset,the deep autoencoder is combined with various classifiers.And the classification effect is compared by experiments.3.In order to develop the feature learning ability of deep autoencoder,this paper proposes an improved algorithm based on particle swarm optimization(PSO)and stack autoencoder(SAE).The main idea of this algorithm is to use the particle swarm optimization algorithm to find the optimal particle which can control the parameters and the stack denoising autoencoders algorithm to extract the feature quickly.First,this algorithm seeks the optimal number of neurons in each layer of the stack autoencoder by the particle swarm optimization algorithm to optimized autoencoder.And then,it extracts features by stack autoencoders.Finally,the KNN algorithm is used for classification.The new algorithm is tested on the MNIST dataset,which verifies the feasibility of the proposed method,accelerates the convergence speed of the whole algorithm,improves the accuracy of classification and enhances the robustness of the whole model.4.This paper proposes a face recognition algorithm based on deep learning,which combines the local binary pattern(LBP)and stack denoising autoencoder(SDAE).LBP can eliminate the influence of illumination and angle,and has the advantages of simple calculation and high efficiency.SDAE has the advantage of quickly extracting features and effectively removing noise.I have done experiments on the YALE face database for proving the new algorithm which improves the accuracy,the robustness and speeds up the calculation.