AI / ML VisualizerRuns in your browser · No API keyOpen Model
← All topics

Representation Learning

Variational Autoencoder

Explore mean, variance, reparameterization and the Gaussian KL term.

Variational autoencoder · distribution to reconstructionVariational autoencoder · distribution to reconstructionInput8 valuesEncoderμ and log σ²Sample z-0.8833, 0.243Decoder8 reconstructions
Input · 1 × 8
Reconstruction · 1 × 8
Mean μ / log variance · 2 × 2
z = μ + exp(½ log σ²) × ε
0.0765 + exp(0.0038) × -0.9562 = -0.8833; 0.5233 + exp(0.0262) × -0.273 = 0.243
MSE 0.2523; KL 0.1406; total 0.2551 · 0 updates
Latent means · click to decode a locationLatent means · click to decode a location-3-3-1.5-1.5001.51.533Point 1: (0.0765, 0.5233), class 0Point 2: (0.4659, -0.4303), class 1Point 3: (0.809, 0.2045), class 2Point 4: (-0.2408, 0.9726), class 3Point 5: (-0.4626, 0.0817), class 4Point 6: (-0.3101, -0.768), class 5Point 7: (-0.1581, -0.4679), class 6Point 8: (0.3499, 0.5673), class 7Point 9: (-0.8833, 0.243), class 8

Continue exploring

Autoencoder →Denoising Autoencoder →Sparse Autoencoder →

No account required. Local images and model files stay in your browser. Educational models demonstrate mechanisms; they are not production predictors.