• Sushree Barsa Pattnayak

Auto encoder-A Deep Learning Approach for Data compression

Introduction to Encoder-

Data compression is a big topic which is used for computer vision, computer networks and many more. The point of data compression is to convert our input into a smaller representation that we recreate to a degree of quality .The smaller representation is what could be past around and when anyone needed the original & reconstructed from the smaller one. Auto-encoders are unsupervised neural networks and used machine learning to do the compression for us. The aim of an auto encoder is to learn the data distribution of given data typically for the purpose of dimensionality reduction.

Auto-encoder is better than PCA –

a. Non-linear transformation

Non linear activation function and multiple layer

b. Convolutional layers

An auto encoder doesn’t have to learn dense layers

c. Higher efficiency

More efficient to learn several layers with auto encoder

d. Multiple transformation

It gives a representation as the output of each layer

Auto encoder neural network is an unsupervised machine learning algorithm that applies back propagation, setting the target values to be equal to the inputs .It provides better result as compared to the PCA and other techniques. It is more similar to PCA but it gives more better result as compared to it. It minimizes the same objective function

Parts of encoder-


This is the part of the network that compresses the input into a latent space representation


This is the part of the network represents the compressed input that is fed to the decoder


This part aims to reconstruct the input the latent space representation

Applications of Auto encoder-

a. Image coloring

b. Feature variation

c. Dimentionality Reduction

d. Denoising Images

e. Watermark Removal