ABSTRACT
In this chapter, we discuss two additional features that can be achieved by a DNN-assisted method: integrity protection and computational efficiency. On integrity, we specifically introduce a notation of quantitative pre-specification to deliver fixed results by a pre-planned method given observed data. This is a higher standard than the traditional qualitative pre-specification to further enforce integrity. We apply this idea to a case study of historical data borrowing for multiple endpoints with a Bayesian hierarchical model. DNNs are able to assist the quantitative pre-specification of the Bayesian posterior inference with Markov Chain Monte Carlo (MCMC). In a similar fashion, DNNs can significantly save computational time in large-scale simulations by spending a moderate training time. This property is demonstrated by assisting the nonparametric Bootstrap method.
