Dynamic Perturbed Adaptive Method for Infinite Task-Conflicting Time Series
This addresses the challenge of continual adaptation in time series analysis for scenarios with frequent task shifts, though it appears incremental as it builds on trunk-branch architectures.
The paper tackles the problem of adapting models to time series tasks with conflicting objectives, where the same input can yield different outputs, by proposing a dynamic perturbed adaptive method that outperforms baselines in complex environments, showing fast adaptation and progressive learning.
We formulate time series tasks as input-output mappings under varying objectives, where the same input may yield different outputs. This challenges a model's generalization and adaptability. To study this, we construct a synthetic dataset with numerous conflicting subtasks to evaluate adaptation under frequent task shifts. Existing static models consistently fail in such settings. We propose a dynamic perturbed adaptive method based on a trunk-branch architecture, where the trunk evolves slowly to capture long-term structure, and branch modules are re-initialized and updated for each task. This enables continual test-time adaptation and cross-task transfer without relying on explicit task labels. Theoretically, we show that this architecture has strictly higher functional expressivity than static models and LoRA. We also establish exponential convergence of branch adaptation under the Polyak-Lojasiewicz condition. Experiments demonstrate that our method significantly outperforms competitive baselines in complex and conflicting task environments, exhibiting fast adaptation and progressive learning capabilities.