Built independently by an author, for readers. Read the story and support ChapterPal
ChapterPal Logo
Log inSign up

keyword

non-autoregressive conditional diffusion models

Non-autoregressive conditional diffusion models are generative neural network frameworks that synthesize entire target sequences or multi-step outputs simultaneously based on given conditioning inputs, rather than predicting outputs one step at a time. In these models, an iterative denoising process gradually transforms random noise into coherent data conditioned on contextual information, such as historical observations or external signals. By generating all sequence steps in parallel rather than sequentially, these models eliminate the compounding error propagation common in step-by-step autoregressive generation, reduce latency during inference, and leverage the expressive density estimation of diffusion processes to capture complex patterns and uncertainties across the entire generated sequence.

1 item

Non-autoregressive Conditional Diffusion Models for Time Series Prediction

Non-autoregressive Conditional Diffusion Models for Time Series Prediction

Lifeng Shen, James T. Kwok

OrganizationsThe Hong Kong University of Science and Technology

Why you should read this

Proposes TimeDiff, a non-autoregressive diffusion framework featuring future mixup and autoregressive initialization to overcome error accumulation and outperform existing transformers and diffusion baselines in long-range time series forecasting.

Recently, denoising diffusion models have led to significant breakthroughs in the generation of images, audio and text. However, it is still an open question on how to adapt their strong modeling ability to model time series. In this paper, we propose TimeDiff, a non-autoregressive diffusion model that achieves high-quality time series prediction with the introduction of two novel conditioning mechanisms: future mixup and autoregressive initialization. Similar to teacher forcing, future mixup allows parts of the ground-truth future predictions for conditioning, while autoregressive initialization helps better initialize the model with basic time series patterns such as short-term trends. Extensive experiments are performed on nine real-world datasets. Results show that TimeDiff consistently outperforms existing time series diffusion models, and also achieves the best overall performance across a variety of the existing strong baselines (including transformers and FiLM).

Added

2026-09-26