Efficient Uncertainty Quantification and Reduction for Over-Parameterized Neural Networks | Read Paper on Bytez