Exogenous Dropout: A Simple, Strong Baseline for Corruption-Robust Time Series Forecasting with Covariates
arXiv:2607.05452v1 Announce Type: new Abstract: Time series forecasters that use exogenous covariates are fragile in deployment: when those covariates are noised, temporally misaligned, or missing, strong exogenous-fusion and exogenous-adapted models can degrade far above the endogenous-only floor. We study whether such robustness requires specialized architectures, or whether it can be obtained through a simple training intervention. We propose exogenous dropout, a model-agnostic method that randomly zeros whole exogenous channels during training. Across electricity-price forecasting, reservo
The proliferation of real-world AI deployments highlights the practical fragility of models facing imperfect data, driving research into robust forecasting methods. This paper emerges now as the demand for reliable AI continually increases across industries.
Improving the robustness of time series forecasting with exogenous variables directly impacts the reliability and deployability of AI systems in critical applications. It mitigates the risk of model failure due to real-world data corruption, enhancing trust and utility.
The proposed 'exogenous dropout' method offers a simple, model-agnostic technique to significantly improve AI model resilience to corrupted input data. This simplifies the development of robust forecasting systems without needing highly specialized architectures.
- · AI developers
- · Industries relying on time series forecasting (e.g., energy, finance)
- · Cloud infrastructure providers
- · Developers relying on brittle, highly specialized forecasting models
Increased reliability and adoption of AI systems in production environments, reducing deployment friction.
Faster development cycles for robust AI solutions as fewer resources are spent on complex data-cleaning pipelines or bespoke architecture design.
More complex and interconnected AI systems become viable, as the foundational robustness of their individual components is enhanced.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG