AdaMamba Framework for Long-Term Time Series Forecasting
AdaMamba is a novel framework for long-term time series forecasting (LTSF) that integrates adaptive frequency-domain analysis with state-space models. The method addresses cross-domain heterogeneity, where time-domain synchronized variables differ in frequency domain. It uses an interactive patch encoding module and frequency-gated Mamba update process to capture complex dependencies and periodic patterns. The approach is validated on multiple real-world datasets, showing improved accuracy over existing methods. The paper is published on arXiv with ID 2604.23239.
Key facts
- AdaMamba is a framework for long-term time series forecasting.
- It integrates frequency-domain analysis with Mamba state-space models.
- It addresses cross-domain heterogeneity in time series data.
- Uses an interactive patch encoding module.
- Employs frequency-gated Mamba update process.
- Validated on real-world datasets.
- Published on arXiv with ID 2604.23239.
- Focuses on capturing long-range dependencies and dynamic periodic patterns.
Entities
Institutions
- arXiv