Bilateral Denoising Diffusion Models
This work addresses the sampling efficiency problem for users of generative models, offering a significant speedup while maintaining or improving sample quality.
The paper tackles the challenge of inefficient sampling in denoising diffusion probabilistic models (DDPMs) by proposing bilateral denoising diffusion models (BDDMs), which achieve high-fidelity samples with as few as 3 steps and a 62x speedup compared to DDPMs using 1000 steps.
Denoising diffusion probabilistic models (DDPMs) have emerged as competitive generative models yet brought challenges to efficient sampling. In this paper, we propose novel bilateral denoising diffusion models (BDDMs), which take significantly fewer steps to generate high-quality samples. From a bilateral modeling objective, BDDMs parameterize the forward and reverse processes with a score network and a scheduling network, respectively. We show that a new lower bound tighter than the standard evidence lower bound can be derived as a surrogate objective for training the two networks. In particular, BDDMs are efficient, simple-to-train, and capable of further improving any pre-trained DDPM by optimizing the inference noise schedules. Our experiments demonstrated that BDDMs can generate high-fidelity samples with as few as 3 sampling steps and produce comparable or even higher quality samples than DDPMs using 1000 steps with only 16 sampling steps (a 62x speedup).