VQ-DRAW: A Sequential Discrete VAE
This work addresses the challenge of discrete representation learning for image data, offering a method that could benefit machine learning applications in compression and synthesis, though it appears incremental as it builds on existing DRAW and VQ techniques.
The paper tackles the problem of learning compact discrete representations for data compression and generation by introducing VQ-DRAW, which adapts a sequential generation scheme to discrete latent variables using vector quantization, resulting in effective image compression and realistic sample generation without an autoregressive prior.
In this paper, I present VQ-DRAW, an algorithm for learning compact discrete representations of data. VQ-DRAW leverages a vector quantization effect to adapt the sequential generation scheme of DRAW to discrete latent variables. I show that VQ-DRAW can effectively learn to compress images from a variety of common datasets, as well as generate realistic samples from these datasets with no help from an autoregressive prior.