CHIME: Conditional hallucination and integrated multi-scale enhancement for time series diffusion model
Yuxuan Chen, Yuxuan Wu, Haipeng Xie · Knowledge-Based Systems · 2026
The denoising diffusion probabilistic model has emerged as a leading generative model, demonstrating significant success in various computer vision tasks. Recently, initial explorations have applied diffusion models to time series tasks; however, existing studies encounter challenges in multi-scale feature alignment and generative capabilities across different entities and few-shot scenarios. This study proposes CHIME, a conditional hallucination and integrated multi-scale enhancement framework for time series diffusion models. By utilizing multi-scale decomposition and integration, CHIME captures the decomposed features of time series, achieving in-domain distribution alignment between generated and original samples. Additionally, we introduce a Feature Hallucination (FH) module in the conditional denoising process to enable generic temporal semantic knowledge transfer. Experimental results on publicly available real-world datasets demonstrate that CHIME achieves state-of-the-art performance and exhibits excellent generalization capabilities in few-shot scenarios