Multi-resolution Time-Series Transformer for Long-term Forecasting
This work addresses the problem of long-term forecasting for applications like finance or weather, but it is incremental as it builds on existing patch-based transformer methods.
The paper tackles the challenge of learning diverse temporal patterns in time-series forecasting by proposing a multi-resolution transformer framework, which achieves state-of-the-art performance on real-world datasets.
The performance of transformers for time-series forecasting has improved significantly. Recent architectures learn complex temporal patterns by segmenting a time-series into patches and using the patches as tokens. The patch size controls the ability of transformers to learn the temporal patterns at different frequencies: shorter patches are effective for learning localized, high-frequency patterns, whereas mining long-term seasonalities and trends requires longer patches. Inspired by this observation, we propose a novel framework, Multi-resolution Time-Series Transformer (MTST), which consists of a multi-branch architecture for simultaneous modeling of diverse temporal patterns at different resolutions. In contrast to many existing time-series transformers, we employ relative positional encoding, which is better suited for extracting periodic components at different scales. Extensive experiments on several real-world datasets demonstrate the effectiveness of MTST in comparison to state-of-the-art forecasting techniques.