Unsupervised denoising feature learning method based on auto-encoder

Abstract

The invention relates to an unsupervised denoising feature learning method based on an auto-encoder, and the method comprises the following steps: adding a noise data layer in front of an input layeron the basis of the auto-encoder, changing an original hidden layer from one layer to three layers, and enabling the dimensions of the three hidden layers to be sequentially decreased progressively; an input original image sequentially passes through the noise data layer, the input layer, the three hidden layers and the output layer to be output to obtain a reconstructed and restored image. According to the method, more discriminative low-dimensional representation can be learned from unmarked high-dimensional image data.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to an unsupervised denoising feature learning method based on an autoencoder. Background technique [0002] With the continuous innovation of Internet and multimedia hardware technology, data often appear high-dimensional. However, the high-dimensionality of data often brings many problems. For example, high-dimensional images greatly increase the space complexity and time complexity of the algorithm. And sometimes it will lead to serious overfitting phenomenon, which makes the model unable to be used in practice. Feature embedding learning is an effective learning method. It not only reduces the dimensionality of data, but also retains most of the physical meaning of the original features. It is suitable for many research fields. However, traditional unsupervised feature embedding learning algorithms also have problems of efficiency and overfitting when faced with high-d...

AI Technical Summary

Problems solved by technology

However, traditional unsupervised feature embedding learning algorithms also have problems of efficiency and overfitting when facing high-dimensional data.

Images