Non-autoregressive Conditional Diffusion Models for Time Series Prediction
This work addresses the challenge of modeling time series for applications in forecasting, offering a novel method that improves performance over strong existing approaches.
The paper tackles the problem of adapting denoising diffusion models for time series prediction by proposing TimeDiff, a non-autoregressive model with novel conditioning mechanisms, and it shows that TimeDiff outperforms existing models and baselines across nine real-world datasets.
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).