DropoutTS: Sample-Adaptive Dropout for Robust Time Series Forecasting

TL;DR

DropoutTS is a versatile, efficient plugin that adaptively adjusts dropout rates based on instance noise levels, significantly improving the robustness of deep time series models against noisy data without changing model architecture.

Contribution

It introduces a novel sample-adaptive dropout mechanism that uses spectral sparsity to dynamically calibrate model capacity based on noise, enhancing robustness in time series forecasting.

Findings

Consistently improves robustness across diverse noise conditions

Achieves superior performance with negligible parameter overhead

Works with various backbone models without architectural changes

Abstract

Deep time series models are vulnerable to noisy data ubiquitous in real-world applications. Existing robustness strategies either prune data or rely on costly prior quantification, failing to balance effectiveness and efficiency. In this paper, we introduce DropoutTS, a model-agnostic plugin that shifts the paradigm from "what" to learn to "how much" to learn. DropoutTS employs a Sample-Adaptive Dropout mechanism: leveraging spectral sparsity to efficiently quantify instance-level noise via reconstruction residuals, it dynamically calibrates model learning capacity by mapping noise to adaptive dropout rates - selectively suppressing spurious fluctuations while preserving fine-grained fidelity. Extensive experiments across diverse noise regimes and open benchmarks show DropoutTS consistently boosts superior backbones' performance, delivering advanced robustness with negligible parameter overhead and no architectural modifications. Our code is available at https://github.com/CityMind-Lab/DropoutTS.