Abstract

Diffusion-based models have shown great promise in molecular generation butoften require a large number of sampling steps to generate valid samples. Inthis paper, we introduce a novel Straight-Line Diffusion Model (SLDM) to tacklethis problem, by formulating the diffusion process to follow a lineartrajectory. The proposed process aligns well with the noise sensitivitycharacteristic of molecular structures and uniformly distributes reconstructioneffort across the generative process, thus enhancing learning efficiency andefficacy. Consequently, SLDM achieves state-of-the-art performance on 3Dmolecule generation benchmarks, delivering a 100-fold improvement in samplingefficiency.

 

Quick Read (beta)