Exformer introduces extreme-adaptive attention for time series forecasting

The Extreme-Adaptive Transformer (Exformer) is a new forecasting framework designed to explicitly model temporal dependencies involving both normal and extreme events in time series data.

• Exformer utilizes an extreme-adaptive attention mechanism composed of three sparse components: Local, Stride, and Extreme.
• The Local and Stride components capture short-term and periodic temporal dependencies, respectively.
• The Extreme component selectively models event-aware dependencies between normal and extreme streamflow patterns.
• Exformer achieves superior 3-day forecasting performance on four real-world hydrologic streamflow datasets compared to state-of-the-art baselines.

The findings demonstrate that explicitly incorporating extreme-aware attention improves the forecasting capacity of Transformer models on imbalanced time series with rare but consequential events.