Coarse and Fine-grained Forecasting Via Gaussian Process Blurring Effect
Abstract: Time series forecasting is a challenging task due to the existence of complex and dynamic temporal dependencies, leading to inaccurate predictions even by the most advanced models. While increasing training data is a common approach to enhance accuracy, it is often a limitted source. In contrast, we are building on successful denoising approaches for image generation by proposing an end-to-end forecast-blur-denoise framework. By training the parameters of the blur model for best end-to-end performance, we advocate for a clear division of tasks between the forecasting and denoising models. This encourages the forecasting model to learn the coarse-grained behavior, while the denoising model is filling in the blurred fine-grained details. Our experiments show that our proposed approach is able to improve the forecasting accuracy of several state-of-the-art forecasting models as well as several other denoising approaches.
Submission Length: Regular submission (no more than 12 pages of main content)
Changes Since Last Submission: The abstract's width exceeded the actual width due to a minor error, but it has since been corrected.
Assigned Action Editor: ~Jeremias_Sulam1
Submission Number: 2221
Loading