Perceive, Route and Modulate: Dynamic Pattern Recalibration for Time Series Forecasting

2026-05-07

AI Summary

This work addresses the limitation of existing deep time series forecasting models, which rely on fixed weights and struggle to adapt to dynamic local temporal patterns, thereby constraining predictive performance. To overcome this, the authors propose a Dynamic Pattern Recalibration (DPR) mechanism that dynamically adjusts hidden states at the token level through a “perceive–route–modulate” pipeline. DPR employs a soft routing distribution to select relevant patterns, generates modulation signals via learnable response bases, and applies lightweight modulation using residual Hadamard products. Designed as a plug-and-play adapter, DPR significantly enhances the performance of diverse backbone architectures. Its standalone variant, DPRNet, achieves competitive results across twelve benchmark datasets, surpassing the constraints of conventional static transformations and enabling temporally aware adaptive modeling.

Abstract

Local temporal patterns in real-world time series continuously shift, rendering globally shared transformations suboptimal. Current deep forecasting models, despite their scale and complexity, rely on fixed weight matrices applied uniformly to all temporal tokens. This creates a static pattern response: models settle into a compromised average, unable to adapt to changing local dynamics. We introduce Dynamic Pattern Recalibration (DPR), a backbone-agnostic mechanism that resolves this via token-level recalibration. Through a lightweight "Perceive-Route-Modulate" pipeline, DPR computes a soft-routing distribution over a learned basis of adaptive response patterns, generating a time-aware modulation vector that recalibrates hidden states via a residual Hadamard product. As a backbone-agnostic adapter, DPR enhances forecasting across diverse architectures with minimal overhead, confirming it addresses a general bottleneck. As a minimalist standalone model, DPRNet achieves competitive performance across 12 benchmarks, validating dynamic recalibration against macroscopic parameter scaling.

Problem

• time series forecasting
• local temporal patterns
• dynamic pattern adaptation
• static pattern response
• temporal token recalibration

Innovation

• Dynamic Pattern Recalibration
• Time Series Forecasting
• Token-level Recalibration
• Adaptive Response Patterns
• Backbone-agnostic Adapter