Exploring uncertainty in regression neural networks for construction of prediction intervals

Deep learning has achieved impressive performance on many tasks in recent years. However, it has been found that it is still not enough for deep neural networks to provide only point estimates. For high-risk tasks, we need to assess the reliability of the model predictions. This requires us to quantify the uncertainty of model prediction and construct prediction intervals. One of the significant advantages of the proposed method is that it simultaneously implements point estimation and uncertainty quantification. In this paper, we explore the uncertainty in regression neural networks to construct the prediction intervals. In general, we comprehensively consider two categories of uncertainties: aleatory uncertainty and epistemic uncertainty. We design a novel loss function, which enables us to learn uncertainty without uncertainty labels. We only need to supervise the learning of regression tasks. In the process of training, the model implicitly learns aleatory uncertainty under the guidance of loss function. And that epistemic uncertainty is accounted for in the ensembled form. Our method correlates the construction of prediction intervals with uncertainty estimation. Experimental results on some publicly available datasets show that the performance of our method is competitive with other state-of-the-art methods.