The Artificial Intelligence Journey — Autoencoders
In general, autoencoders are a specific type of a neural network (https://medium.com/@boutnaru/introduction-to-neural-networks-f65bf17afd2c) which can be used for dimension reduction by learning an efficient coding of unlabeled data. We can say it is trained to compress the input data (to a lower dimension) and then reconstructing the data in order to match it to the original input as closely are possible (https://deepai.org/machine-learning-glossary-and-terms/autoencoder).
Thus, the architecture of an autoencoder is composed of three layers: encoder, code (aka the bottleneck) and decoder — as shown in the diagram below (https://stackabuse.com/autoencoders-for-image-reconstruction-in-python-and-keras/). The encoder layer compresses the input data into a latent-space representation. The code layer represents the compressed input fed to the decoder layer. The decoder layer reconstructs the input data from the latent space representation (https://www.simplilearn.com/tutorials/deep-learning-tutorial/what-are-autoencoders-in-deep-learning).
Moreover, the goal is to train autoencoders to minimize the reconstruction error. This is done by using a loss/fitness/error function (https://medium.com/@boutnaru/the-artificial-intelligence-journey-loss-function-3534db6c0d69) an example is MSE (mean squared error) between the input data and the reconstructed data. The weights of the neural network are adjusted during the training phase to minimize the loss function (https://www.v7labs.com/blog/autoencoders-guide) mostly using backpropagation.
Lastly, there are different types of autoencoders such as: sparse autoencoders, contractive autoencoders, undercomplete autoencoders, denoising autoencoders and variational autoencoders (https://www.v7labs.com/blog/autoencoders-guide). See you in my next writeup ;-) You can follow me on twitter — @boutnaru (https://twitter.com/boutnaru). Also, you can read my other writeups on medium — https://medium.com/@boutnaru. You can find my free eBooks at https://TheLearningJourneyEbooks.com.
