WAVE: Weighted Autoregressive Varying Gate for Time Series Forecasting
PDF Proceedings arXiv Code Poster
Adds ARMA structure to autoregressive attention via a weighted varying gate, decoupling long‑range and local effects and improving TSF quality without increasing asymptotic complexity.
Demonstrates that properly tokenized autoregressive decoder‑only Transformers are strong TSF baselines and that introducing AR and MA components through WAVE yields consistent accuracy gains across efficient attentions.