Online Residual Correction for Frozen Models

Auto-generated stub. Edit this file to add more details.

A technique to maintain the accuracy of a frozen, pre-trained time-series model by applying real-time corrections based on the residuals (errors) observed in recent predictions.

Why It Matters

This technique adapts a frozen forecasting model to changing conditions by applying corrections based on recent errors, offering a trade-off between retraining cost and accuracy maintenance.

Evidence

In the frozen-model regime, online residual correction improved Facebook Prophet and SARIMAX, with corrected SARIMAX yielding the lowest overall error

Related Papers

• openalex-2603.25495-interpretable-pm25-forecasting-for-urban-air-quality-a-compa