🔄 Autoencoder Visualization

Encoder

Input: 784
Hidden: 256
Latent: 32
→

Decoder

Latent: 32
Hidden: 256
Output: 784

Autoencoder Networks

Architecture: Encoder compresses input to latent representation, decoder reconstructs from latent code

Training: Minimize reconstruction error. Learn meaningful compressed representations.

Latent Space: Low-dimensional representation capturing essential features

Types: Vanilla • Variational (VAE) • Denoising • Sparse • Convolutional

Applications: Dimensionality reduction • Denoising • Anomaly detection • Data compression • Feature learning