Uncertainty Prediction for Deep Sequential Regression Using Meta Models
This work makes sequential regression more effective and practical for real-world applications by improving uncertainty quantification in non-stationary environments.
The paper tackles the challenging problem of generating high-quality uncertainty estimates for deep sequential regression, particularly for recurrent networks in non-stationary real-world scenarios. It presents a flexible method that outperforms competitive baselines on both drift and non-drift scenarios, producing symmetric and asymmetric uncertainty estimates without stationarity assumptions.
Generating high quality uncertainty estimates for sequential regression, particularly deep recurrent networks, remains a challenging and open problem. Existing approaches often make restrictive assumptions (such as stationarity) yet still perform poorly in practice, particularly in presence of real world non-stationary signals and drift. This paper describes a flexible method that can generate symmetric and asymmetric uncertainty estimates, makes no assumptions about stationarity, and outperforms competitive baselines on both drift and non drift scenarios. This work helps make sequential regression more effective and practical for use in real-world applications, and is a powerful new addition to the modeling toolbox for sequential uncertainty quantification in general.