FreDF: Learning to Forecast in the Frequency Domain

2024-02-04

AI Summary

Existing direct forecasting (DF) paradigms neglect the autocorrelation among multi-step future labels, leading to biased learning objectives. To address this, this paper introduces— for the first time—frequency-domain modeling into the DF framework, proposing a general, plug-and-play frequency-domain calibration method. Specifically, the prediction task is mapped to the frequency domain via the Discrete Fourier Transform (DFT), where label-wise correlations are explicitly modeled to mitigate estimation bias and rectify the learning objective. The method seamlessly integrates with state-of-the-art models such as Informer and Autoformer. Extensive experiments on multiple benchmark datasets demonstrate an average MAE reduction of over 12% compared to current SOTA methods, yielding substantial improvements in forecasting accuracy. The approach exhibits strong generality and scalability across diverse architectures and datasets.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Planning, Routing, and Scheduling: Planning with Language Models
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Time series modeling presents unique challenges due to autocorrelation in both historical data and future sequences. While current research predominantly addresses autocorrelation within historical data, the correlations among future labels are often overlooked. Specifically, modern forecasting models primarily adhere to the Direct Forecast (DF) paradigm, generating multi-step forecasts independently and disregarding label autocorrelation over time. In this work, we demonstrate that the learning objective of DF is biased in the presence of label autocorrelation. To address this issue, we propose the Frequency-enhanced Direct Forecast (FreDF), which mitigates label autocorrelation by learning to forecast in the frequency domain, thereby reducing estimation bias. Our experiments show that FreDF significantly outperforms existing state-of-the-art methods and is compatible with a variety of forecast models. Code is available at https://github.com/Master-PLC/FreDF.

Problem

• Addresses autocorrelation in future time series labels
• Mitigates bias in Direct Forecast learning objectives
• Enhances forecasting via frequency domain learning

Innovation

• Forecasts in frequency domain to reduce bias
• Mitigates label autocorrelation via frequency learning
• Compatible with diverse forecasting models