We initiate a systematic study of algorithms that are both differentially-private and run in sublinear time for several problems in which the goal is to estimate natural graph parameters. Our main result is a differentiallyprivate (1 + ρ)-approximation algorithm for the problem of computing the average degree of a graph, for every ρ > 0. The running time of the algorithm is roughly the same as its non-private version proposed by Goldreich and Ron (Sublinear Algorithms, 2005). We also obtain the first differentially-private sublinear-time approximation algorithms for the maximum matching size and the minimum vertex cover size of a graph.An overarching technique we employ is the notion of coupled global sensitivity of randomized algorithms. Related variants of this notion of sensitivity have been used in the literature in ad-hoc ways. Here we formalize the notion and develop it as a unifying framework for privacy analysis of randomized approximation algorithms. J. B. and T.M were supported in part by NSF CNS-1931443 and NSF CCF-1910659. E.G and T. M. were supported in part by NSF CCF-1910659 and NSF CCF-1910411. 1 GraphsGraphs G 1 and G 2 are edge-neighboring i.e., G 1 ∼ e G 2 if there exists an edge e such that E 1 \ {e} = E 2 \ {e}. 1 algorithms. They showed how to estimate the cost of a minimum spanning tree and the number of triangles in a graph by calibrating noise to a local variant of sensitivity called smooth sensitivity. Subsequent works in designing edge differentially-private algorithms for computing graph statistics include [16,15,18,31]. Gupta, Ligett, McSherry, Roth and Talwar [14] gave the first edge differentially-private algorithms for classical graph optimization problems, such as vertex cover, and minimum s-t cut, by making clever use of the exponential mechanism in existing non-private algorithms that solve the same problem.An even more desirable notion of privacy in graphs is the notion of node differential privacy i.e., neighboring graphs that differ by a single node and edges incident to it in Definition 1. The concept of node differentiallyprivate algorithms for 1-dimensional functions (functions that output a single real value) on graphs was first rigorously studied independently by Kasiviswanathan, Nissim, Raskhodnikova and Smith [17], as well as, Blocki, Blum, Datta, and Sheffet [2], and Chen and Zhou [6]. Their techniques were later extended to higher-dimensional functions on graphs [23,3]. Subsequent works have focused on developing node differentially-private algorithms for a family of network models: stochastic block models and graphons [4,25]. A more recent line of work has focused on the continual release of graph statistics such as degree-distributions and subgraph counts in an online setting [28,11]. Gehrke, Lui, and Pass [12] introduce a more robust notion of differential privacy called Zero-Knowledge Differential Privacy (ZKDP), which tackles the problem of auxiliary information in social networks. This work uses existing results from sublinear-time algorithms as a building...