Recovering Network Structures Based on Evolutionary Game Dynamics via Secure Dimensional Reduction

Abstract

The curse of dimensionality is a challenging issue in network science: the problem of inferring the network structure from sparse and noisy data becomes more and more difficult, indeed, as their dimensionality increases. We here develop a general strategy for dimensional reduction using iteratively thresholded ridge regression screener, one statistical method aiming to resolve the problem of variable selection. After drastically reducing the dimensions of the problem, we then employ the lasso method, a convex optimization method, to recover the network structure. We demonstrate the efficiency of the dimensional reduction method, and particular suitability for the natural sparsity of complex networks, in which the average degree is much smaller than their total number of nodes. Analysis based on various game dynamics and network topologies show that higher reconstruction accuracies and smaller reconstruction times can be achieved by our method. Our approach provides, therefore, a novel insight to solve the reconstruction problem and has potential applications in a wide range of fields.