drah-kah-ris/auto_encoders

Autoencoders using TensorFlow

Using TensorFlow creating an autoencoder model to reduce the dimensionality and comparing the reconstruction loss bewteen neural networks and PCA.

Please check the notebook: https://github.com/drah-kah-ris/auto_encoders/blob/master/tf_autoencoders.ipynb

It is evident from the results in the above notebook that AutoEncoders would produce similar results as that of PCA's when the activation function is linear irrespective of the number of hidden layers. When the activation function is non-linear and the hidden layers are more Auto Encoders beats the PCA's results.