predict-then-correct

The paper proposes a Predict-then-Correct (PtC) framework using a few-shot continuous contextual bandit to adaptively correct base ML forecasts in retail demand forecasting, achieving significant improvements in error metrics and inventory costs over baselines.