The new contender to GANs: score matching with Langevin SamplingLast year, Yang Song, a graduate student at Stanford, showed an entirely new way of generating data based on denoising score matching with Annealed Langevin sampling (DSM-ALS). The paper showed that a non-adversarial approach could reach levels similar to GANs (FID of 25, which is what Relativistic GANs reached two years ago). Surprisingly, the paper received little to no attention in the research
