Neural Distributed Source Coding

Explore a groundbreaking framework for lossy Distributed Source Coding (DSC) in this 24-minute lecture by Hyeji Kim from The University of Texas at Austin. Delve into a novel approach that overcomes limitations of traditional DSC methods by utilizing a conditional Vector-Quantized Variational AutoEncoder (VQ-VAE) to learn distributed encoders and decoders. Discover how this technique achieves state-of-the-art PSNR while handling complex correlations and scaling to high dimensions, all without relying on hand-crafted source modeling. Gain insights into the application of this method across multiple datasets and understand its potential to revolutionize practical DSC implementation beyond synthetic datasets and specific correlation structures.