Researchers have developed a new method called Endpoint-Preserving Online Correction (EPOC) for multi-horizon time series forecasting. EPOC addresses the challenge of retaining residual feedback without excessive state storage by compressing the residual state using low-order discrete cosine transform (DCT) coefficients and the preceding residual block's final value. Evaluations on multivariate series with DLinear and PatchTST models demonstrated that EPOC significantly reduces mean squared error (MSE) and mean absolute error (MAE) compared to uncorrected base forecasts, while using substantially less auxiliary state than other methods like the $\delta$-Adapter and COSA. AI
IMPACT This research introduces a novel technique for improving time series forecasting accuracy and efficiency by compressing residual states, potentially benefiting applications requiring precise future predictions.
RANK_REASON This is a research paper detailing a new method for time series forecasting. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
