CAMP: Cycle-Aware Multi-Scale Patch Mixer Achieves State-of-the-Art on Six Long-Term Forecasting Benchmarks
Summary:A new arXiv paper introduces CAMP, a Cycle-Aware Multi-Scale Patch Mixer for time series forecasting. CAMP addresses limitations of existing cycle-aware methods by adaptively learning dominant frequencies per input window without predefined cycle lengths, using a horizon-guided patch mixer for position-dependent refinement, and modeling de-cycled residuals via temporally aligned multi-resolution representations. Across seven long-term forecasting benchmarks, CAMP achieves the best average MSE on
Overview
Time series forecasting is critical in finance, transportation, energy, and other domains. Real-world time series are often governed by recurring patterns, but their dominant periods may vary across datasets, forecasting settings, and individual input windows. Existing cycle-aware forecasters typically rely on a single period selected at the dataset level, which can be restrictive when periodic behavior changes over time or when multiple cycles coexist. Moreover, patch-based models usually process all patch positions uniformly, although patches farther from the forecast boundary may require broader contextual refinement, while recent patches contain information that should be preserved more directly. After cyclic behavior is removed, the remaining dynamics may also span multiple temporal resolutions and cannot be adequately described at a single scale.
To address these challenges, a new arXiv paper introduces CAMP (Cycle-Aware Multi-Scale Patch Mixer).
Key Innovations
Adaptive Cycle Learning Module
CAMP's adaptive cycle learning module identifies dominant frequencies separately for each input window and generates both historical and future cyclic components without requiring a predefined cycle length. This enables the model to adapt to time-varying periodic behavior and handle coexisting multiple cycles.
Horizon-Guided Patch Mixer
CAMP introduces a horizon-guided patch mixer that enables position-dependent refinement: earlier patches can incorporate broader temporal context while preserving information close to the forecast boundary. This addresses the limitation of uniform processing in traditional patch models.
Multi-Scale Residual Modeling
CAMP models the de-cycled residual through temporally aligned multi-resolution representations, allowing complementary dynamics at different scales to be captured within one forecasting framework.
Experimental Results
On seven long-term forecasting benchmarks, CAMP achieves remarkable results:
- Best average MSE on six datasets;
- Best or tied-best MAE on six datasets;
- Highest MSE win count across sixteen settings on four PEMS traffic benchmarks.
These results demonstrate CAMP's effectiveness and robustness in cycle-aware forecasting.
Industry Impact and Developer Recommendations
CAMP offers a new paradigm for time series forecasting, especially in scenarios with complex and varying periodic patterns. Developers can leverage its ideas of adaptive cycle learning, position-dependent patch mixing, and multi-scale residual modeling to improve existing forecasting models. This work also provides new benchmarks and directions for future research.
Reference
- Paper link: https://arxiv.org/abs/2608.04051
Tags: #Time series projections #Periodic awareness #Multiscale Modelling #CAMP #In-depth learning
Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.
Community Comments