Prompt Description

Figure Title: Framework of M=2 Co-teaching Training for Noisy Time Series Forecasting Overall Layout: A left-to-right pipeline featuring two parallel model branches at its center. The framework comprises four vertical stages: Input, Parallel Prediction, Sample Selection, and Cross Update. 1. Input Module (Left): A single block on the far left is labeled "Input Time Series Windows." The block contains "Sliding window segments" with a length of *L* and a prediction horizon of *H*. An annotation below states: "Noisy supervision is injected into prediction targets; input sequences remain clean." A single arrow from this block splits into two parallel paths, directed towards Model 1 and Model 2. 2. Parallel Model Branches (Center): Two identical model blocks are positioned side by side. Model Blocks: The upper block is labeled "Model *f*θ1". The lower block is labeled "Model *f*θ2". Both blocks contain the annotation: "Same architecture, independent parameters." Each model block receives input from the Input Time Series Windows.