A Comprehensive Guide to Variational Autoencoders (VAEs)

Adding stochasticity to conventional autoencoders

Rukshan Pramoditha
8 min readJul 1, 2024
Photo by Nat on Unsplash

Conventional (traditional) autoencoders

Autoencoders are a type of neural network architecture that belongs to unsupervised learning.

There are different variants of autoencoders, and they can be classified in many different ways using various criteria.

Previously, I have discussed the conventional (traditional) autoencoder architecture with some practical applications such as image generation, image compression/dimensionality reduction, image denoising, and image colorization.

As a quick recap, a conventional autoencoder consists of an Encoder, Decoder, and Latent (Compressed/Compact/Encoded) Representation (Vector/Space/Code).

The following image shows the encoding and decoding process of a conventional autoencoder.

The encoding and decoding process of a traditional AE (Image by author)

The encoder function, E takes the input, X, and transforms it into a compressed representation called Z.

X = Input
E = Encoder function (non-linear)
Z = Compressed (encoded)…

--

--

Rukshan Pramoditha

3,000,000+ Views | BSc in Stats | Top 50 Data Science, AI/ML Technical Writer on Medium | Data Science Masterclass: https://datasciencemasterclass.substack.com