Autoencoder
A neural network trained to compress input data into a compact representation (encoding) and then reconstruct the original data from that representation.
Architecture
An autoencoder has two parts: an encoder that compresses input into a lower-dimensional latent space, and a decoder that reconstructs the input from that compressed representation. The network learns by minimizing reconstruction error.
Variants
Variational Autoencoder (VAE): Adds a probabilistic twist, learning a distribution over the latent space. Enables generation of new data. Denoising Autoencoder: Trained to reconstruct clean data from corrupted inputs. Sparse Autoencoder: Adds a sparsity constraint to learn more meaningful features.
Applications
Dimensionality reduction, image denoising, anomaly detection, feature learning, and as components in generative models.