DINOISER: Diffused Conditional Sequence Learning by Manipulating Noises

arXiv:2302.10025v2 Announce Type: replace-cross
Abstract: While diffusion models have achieved great success in generating continuous signals such as images and audio, it remains elusive for diffusion models in learning discrete sequence data like natural languages. Although recent advances circumvent this challenge of discreteness by embedding discrete tokens as continuous surrogates, they still fall short of satisfactory generation quality. To understand this, we first dive deep into the denoised training protocol of diffusion-based sequence generative models and determine their three severe …

abstract advances arxiv audio challenge continuous cs.ai cs.cl cs.lg data diffusion diffusion models embedding images languages natural success tokens type while