CARE: A Lightweight Plug-in Gated Correction and Uncertainty-aware Module for Long-term Time Series Forecasting
CARE is a lightweight plug-in added to deterministic forecasters to improve long-term time series forecasting by learning residual correction patterns. It works by analyzing aligned historical context and applying scale-aware, bounded updates modulated by sigmoid risk gates. This allows the correction operator to evolve jointly with the main model, focusing on structures the base forecaster misses.