“…One of the key challenges is that directly applying statistical meth-ods to super-large data with conventional computing approaches is prohibitive, which calls for the development of new tools. Recently, statistical analysis and inference in the large-scale dataset have drawn much attention, and some computationally scalable methods have been proposed to reduce the computation and storage effort from various aspects of applications, such as the divide-and-conquer procedures (Battey et al, 2018;Jordan et al, 2019;Zhao et al, 2019), subsampling strategies (Kleiner et al, 2014;Wang et al, 2018) and on-line learning methods (Balakrishnan and Madigan, 2008;Schifano et al, 2016). In most of these works, one usually assumes a parametric model, typically linear or logistic regression models.…”