Researchers propose GatedLinear, a lightweight framework that addresses the limitations of single-mechanism deep learning models by framing time series forecasting as the adaptive routing of complementary linear bases. The system utilizes a pool of three specialized mechanisms: a global trend-seasonal basis, a difference-based incremental basis for nonstationary drift, and a phase-aligned recurrence basis for cyclic reuse.
- A Tri-Factorized Fusion Gate disentangles routing decisions into channel-specific preferences, horizon-aware offsets, and phase-indexed biases derived from known future time marks.
- The design enables highly granular, point-wise soft routing across different predictive regimes without stacking computationally heavy neural modules.
- Experiments on standard benchmarks show the method achieves state-of-the-art or highly competitive accuracy against recent complex foundational models.
The approach offers explicitly interpretable routing patterns while operating with a substantially smaller parameter footprint.