RCCP: Retrieval-Corrected Conformal Prediction for Time Series
A novel approach for predicting time series intervals, named Retrieval-Corrected Conformal Prediction (RCCP), has been introduced in a paper on arXiv (2608.10553). This technique tackles the shortcomings of traditional conformal prediction (CP) calibration for time series data, where forecast inaccuracies are influenced by time and conditions. While recent CP advancements enhance local calibration through recent, weighted, or localized residuals, local calibration may still be indirect due to extensive residual weighting or extra adaptation steps. RCCP utilizes comparable past residuals as local evidence and rectifies coverage errors caused by retrieval. It formulates an asymmetric interval from the retrieved one-sided residuals. The full paper can be accessed at https://arxiv.org/abs/2608.10553.
Key facts
- The paper proposes Retrieval-Corrected Conformal Prediction (RCCP) for time series prediction intervals.
- RCCP is a retrieval-augmented calibration method.
- It addresses inefficiencies in standard conformal prediction for time series data.
- RCCP selects similar past residuals as local evidence.
- It corrects coverage errors left by retrieval.
- RCCP builds an asymmetric interval from retrieved one-sided residuals.
- The paper is on arXiv with ID 2608.10553.
- The source URL is https://arxiv.org/abs/2608.10553.
Entities
Institutions
- arXiv