GatedLinear: Adaptive Routing of Complementary Linear Bases for Time Series Forecasting

2026-07-10 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

GatedLinear is a lightweight framework designed for time series forecasting, addressing the challenge of diverse temporal dynamics that often overwhelm single-mechanism deep learning models. It redefines forecasting as the adaptive routing of complementary linear bases, utilizing a pool of three specialized mechanisms: a global trend-seasonal basis for smooth projections, a difference-based incremental basis for nonstationary drift, and a phase-aligned recurrence basis for cyclic reuse. A Tri-Factorized Fusion Gate dynamically orchestrates these mechanisms by disentangling routing decisions into channel-specific preferences, horizon-aware offsets, and phase-indexed biases. This approach enables granular, point-wise soft routing, achieving state-of-the-art or highly competitive accuracy on standard benchmarks with a substantially smaller parameter footprint and interpretable routing patterns.

Key takeaway

For Machine Learning Engineers building time series forecasting systems, GatedLinear offers a compelling alternative to complex foundational models. You should consider implementing adaptive routing with specialized linear bases to handle diverse temporal dynamics more effectively. This approach can yield state-of-the-art accuracy with significantly fewer parameters, improving both performance and interpretability in your predictive models.

Key insights

GatedLinear adaptively routes specialized linear bases via a Tri-Factorized Fusion Gate to forecast diverse time series dynamics efficiently.

Principles

• Diverse temporal dynamics require specialized predictive treatments.
• Adaptive routing of complementary bases improves forecasting accuracy.
• Disentangling routing decisions enhances model interpretability.

Method

GatedLinear employs a Tri-Factorized Fusion Gate to dynamically route inputs across three specialized linear bases: trend-seasonal, incremental, and phase-aligned recurrence, enabling point-wise soft routing.

In practice

• Apply adaptive routing for heterogeneous time series data.
• Utilize specialized bases for distinct temporal patterns.
• Design gates for granular, interpretable routing decisions.

Topics

• Time Series Forecasting
• Adaptive Routing
• Linear Models
• Gated Networks
• Model Interpretability
• Parameter Efficiency

Best for: Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• TimeRouter: Efficient and Adaptive Routing of Time-Series Foundation Models — TimeRouter efficiently routes diverse time-series foundation models using discriminative routing and selective gating, avoiding LLM inference overhead.
• RhyMix: A Lightweight Adaptive Multi-Rhythm Network for Long-Term Time Series Forecasting — RhyMix uses a dual-path, adaptively gated architecture to capture diverse temporal patterns in long-term time series forecasting.
• Route, Communicate, and Reason: Gated Routing and Adaptive Depth for Efficient Multi-Agent Reasoning — GRADE optimizes multi-agent reasoning via gated routing and adaptive depth, reducing compute while improving performance.
• Hopformer: Homogeneity-Pursuit Transformer for Time Series Forecasting — Hopformer unifies common temporal patterns in multiple time series while retaining series-specific information.
• Efficient and Trainable Language Model Test-Time Scaling via Local Branch Routing — Local Branch Routing (LBR) enhances language model reasoning by efficiently exploring local token futures via a trainable router.
• How Good Can Linear Models Be for Time-Series Forecasting? — Tuning preprocessing for linear models can achieve competitive time-series forecasting accuracy against larger deep learning architectures.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.