52. VQ-VAE

Learning objectives

• Introduce VQ-VAE

• You can add notes on each slide with blocks like this!
• Load a deck in the browser and type “s” to see these notes.

Variational Auto-Encoders

Variational Auto-Encoders

In AI architecture, variational autoencoders simulate the latent space (between the encoder and decoder)—usually with a mixture of Gaussian functions—to maximize the ELBO (evidence lower bound)

VAE Gaussians

• image source

without

without VAE

• image source: Jeff Orchard

with

with VAE

• image source: Jeff Orchard

VQ-VAE

VQ-VAE

Vector quantized variational autoencoders (VQ-VAE) utilize an discrete embedding space

• for example: 32x32 embedding space of vectors

VQ-VAE

• image source