Inferring infection rate based on observations in complex networks

The infection rate of a propagation model is an important factor for characterizing a dynamic diffusion process accurately, which determines the scale and speed of a diffusion. Inferring an infection rate, based on an observed propagation phenomenon, can help us better estimate the threat of a diffusion in advance and deploy corresponding strategies to restrain such diffusion. Meanwhile, the infection rate is a vital and predefined parameter in the field of propagation network reconstruction and propagation source identification. Therefore, how to infer an infection rate effectively from observed diffusion data is of great significance. In this paper, a backpropagation-based maximum likelihood estimation (BP-ML) is used to infer such infection rate. More specifically, a set of sensors are first deployed into a network for collecting diffusion data (i.e., the infection time of a node). Then, a series of backpropagations are initiated by nodes resided by these sensors in order to deduce the more probable infection rate based on the maximum likelihood estimation. Some experiments in real-world networks show that by taking full advantage of observed diffusion data, our proposed method can infer the infection rate of a diffusion accurately.