Sawtooth Sampling for Time Series Denoising Diffusion Implicit Models

TL;DR

This paper introduces Sawtooth Sampling, a novel method that accelerates diffusion-based time series generation by 30 times while improving sequence quality for classification, applicable to any pretrained diffusion model.

Contribution

The paper proposes the Sawtooth Sampler, a new technique that significantly speeds up diffusion model sampling and enhances generated sequence quality for time series data.

Findings

Achieves 30x faster sampling speed.

Improves quality of generated sequences for classification.

Applicable to any pretrained diffusion model.

Abstract

Denoising Diffusion Probabilistic Models (DDPMs) can generate synthetic timeseries data to help improve the performance of a classifier, but their sampling process is computationally expensive. We address this by combining implicit diffusion models with a novel Sawtooth Sampler that accelerates the reverse process and can be applied to any pretrained diffusion model. Our approach achieves a 30 times speed-up over the standard baseline while also enhancing the quality of the generated sequences for classification tasks.