Towards Robust Time Series Learning via Capacity-Centric Modulation

2026-09-30

AI Summary

This study addresses the heterogeneity of sample reliability in time series learning and the tendency of uniform regularization to induce overfitting or underfitting. To overcome these limitations, this work proposes a Capacity-Centric Modulation framework that introduces a novel sample-adaptive capacity modulation principle. By leveraging spectral sparsity analysis to examine internal activation pathways, the method achieves fine-grained, sample-level allocation of dropout probabilities. The framework operates as a plug-and-play module without requiring architectural redesigns, while preserving deterministic inference and incurring zero overhead during testing. Extensive experiments across 301 datasets demonstrate substantial improvements: the mean squared error for forecasting is reduced by an average of 6.7%, classification accuracy increases by 3.04%, and the F1 score for anomaly detection improves by 17.05%.

Abstract

Sample-level reliability heterogeneity is common in deep time series learning. Standard training pipelines apply a uniform regularization setting to all samples, which can under-regularize corrupted samples and over-restrict clean samples. Common robustness approaches filter observations in data space or impose priors on latent representations. We propose Capacity-Centric Modulation (CCM) as a complementary, sample-adaptive regularization principle. Under this principle, we introduce SACM (Sample-Adaptive Capacity Modulation), a task-agnostic framework that exploits spectral sparsity to assign sample-wise dropout probabilities along internal activation paths. SACM integrates into existing backbones without architectural redesign and preserves the deterministic inference pipeline. Across 301 real-world dataset-backbone pairs covering 9 forecasting, 32 classification, and 4 anomaly-detection datasets, SACM reduces forecasting MSE by 6.7% on average and improves classification accuracy and point-adjusted F1 by 3.04% and 17.05%, respectively, relative to unmodified backbones, with zero test-time overhead.

Problem

• Time Series Learning
• Robustness
• Sample-level Reliability Heterogeneity
• Regularization

Innovation

• Capacity-Centric Modulation
• Sample-Adaptive Regularization
• Spectral Sparsity
• Time Series Learning
• Task-Agnostic Framework