|
Abstract:
|
While the use of network analysis has now permeated most domains, an overwhelming proportion of network analysis methods still work as if the networks we observe are noise free. In many settings, such an assumption could not be further from the truth. Examples include most biological networks, connectomes in neuroscience, contact networks in epidemiology, and, in fact, many networks used in social media studies. In this talk, I will survey a handful of projects from our group and collaborators in recent years that provide simple methods-of-moments approaches to network analysis tasks in a number of settings, which allow users to obtain unbiased inferences of network-related parameters under 'noisy networks'. These estimators are accompanied by confidence intervals deriving from a novel bootstrap algorithm. I will illustrate with application to counting subgraphs, controling epidemic spread, and quantifying treatment effects in network experiments.
|
