DropoutTS: Sample-Adaptive Dropout for Robust Time Series Forecasting

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 DropoutIS, a model-agnostic plugın 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. Code is available at https://github.com/CityMind-Lab/DropoutTS.

Time series data like temperature readings and energy usage drives many important decisions, but forecasting models are fragile. Real world measurements are often corrupted by sensor glitches or transmission errors, leading to unreliable predictions. Existing solutions either discard potentially useful data or require expensive upfront calculations. Our approach shifts from what to learn to how much to learn from each example. We automatically evaluate the reliability of every data point by analyzing its internal structure. Noisy data is given less influence during training, while clean data is fully utilized, letting the model ignore random fluctuations while preserving real patterns. Our method is lightweight and works with existing models without changing their design. Experiments across various noisy scenarios show it consistently improves prediction accuracy with minimal extra cost.