Reliable neural networks for regression uncertainty estimation

While deep neural networks are highly performant and successful in a wide range of real-world problems, estimating their predictive uncertainty remains challenging task. To address this challenge, we propose implement loss function for regression estimation based on the Bayesian Validation Metric (BVM) framework while using ensemble learning. The proposed reproduces maximum likelihood limiting case. A series experiments in-distribution data show that method is competitive with existing state-of-the-art methods. Experiments out-of-distribution robust to statistical change exhibits superior capability.