🤖 AI Summary
To address the low sampling efficiency and high computational cost of diffusion models in time-series synthesis, this paper proposes a general-purpose acceleration framework based on implicit diffusion modeling. We introduce a novel Sawtooth sampler that accelerates the denoising process without modifying pre-trained models. The sampler employs non-uniform step-size scheduling and gradient-guided trajectory optimization to reduce the number of iterations while preserving generation fidelity. Evaluated on standard time-series benchmarks, our method achieves an average 30× speedup over vanilla DDPM, with generated samples exhibiting superior statistical similarity and higher downstream classification accuracy compared to both DDPM and state-of-the-art acceleration baselines. Our key contribution lies in the first integration of implicit diffusion with structured sampling strategies—achieving a favorable trade-off among inference efficiency, sample quality, and model-agnostic applicability.
📝 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.