This work focuses on development of new tools for quantitative accuracy estimation to broaden the applicability of DNNs and related simulation tools from error-tolerant applications (like social media and speech) to error-averse ones (predictive science).