EngineersofAI LogoCoursesBlogResearch LabAI LettersThe LabCode BankInteractive 3DKodrEarnestJobs
Skip to main content
Interactive 3D/Autoencoder Latent Space
Ready
Autoencoder
Encode to latent space, decode back.
Input Digit
Parameters
Latent dim2
2 (compact)16
Noise (latent)0.00
Actions
HUD
Reconstruction loss
0.0750
Latent z
-
Compression ratio
32×
How It Works
The encoder compresses input (64 dims) down to a small latent vector (z). The decoder reconstructs from z.

Interpolation: walk from one class center to another in latent space - the decoder shows a smooth transition.

Noise on the latent vector simulates a VAE's stochastic bottleneck.

Autoencoder Latent Space - Interactive Visualization

An autoencoder trains an encoder to compress input data into a low-dimensional latent vector and a decoder to reconstruct the original input from that vector. The bottleneck forces the model to learn the most information-dense representation of the data. By exploring the latent space, you can interpolate between data points and discover the structure the model has learned to encode.

  • Visualize the encoder architecture: input shrinks layer by layer to the bottleneck latent code
  • Plot the 2D latent space and see how different classes cluster - the geometry the encoder has learned
  • Drag a point in latent space and watch the decoder generate the corresponding reconstruction in real time
  • Understand reconstruction loss: the MSE between input and output drives the encoder to preserve essential information
  • Compare regular autoencoder vs VAE: the VAE adds a KL divergence term that forces a smooth, continuous latent space

Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.