Prompt Description

M = 2 Co-teaching Framework for Noisy Time Series Forecasting The framework employs a sequential, left-to-right pipeline with two parallel model branches, comprising four stages: Input, Parallel Prediction, Sample Selection, and Cross Update. Input Module The input consists of sliding-window time series segments with a window length of L and a prediction horizon of H. Noise is introduced exclusively into the prediction targets, while the input sequences remain noise-free. The identical input windows are simultaneously fed into two parallel models. Parallel Model Branches Two models, f(θ1) and f(θ2), are instantiated with identical architectures but independent parameters. Each model processes the same input windows in parallel and generates its own prediction. Prediction and Window-level Loss Each model outputs a prediction, denoted as Ŷ(1) or Ŷ(2). A window-level loss is computed for each input window by aggregating the prediction errors over the forecasting horizon, utilizing loss functions.