Optimal uncertainty-guided neural network training

The neural network (NN)-based direct uncertainty quantification (UQ) methods have achieved the state of art performance since first inauguration, known as lower–upper-bound estimation (LUBE) method. However, currently-available cost functions for guided NN training are not always converging, and all converged NNs do generate optimized prediction intervals (PIs). In recent years researchers proposed different quality criteria PIs that raise a question about their relative effectiveness. Most existing customizable, convergence is uncertain. Therefore, in this paper, we propose highly customizable smooth function developing to construct optimal PIs. method computes average width PIs, PI-failure distances, PI coverage probability (PICP) test dataset. We examine wind power generation, electricity demand, temperature forecast datasets. Results show reduces variation accelerates training, improves from 99.2% 99.8%. • Proposed training. NN-training becomes faster, higher. customizable. Present philosophies techniques. demand.