A/B Testing in Networks with Adversarial Members
published: Dec. 1, 2017, recorded: August 2017, views: 744
Report a problem or upload filesIf you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.
Many researchers attempt to study the effects of interventions in network systems. To simplify experimental design and analysis in these environments, simple assumptions are made about the behavior of its members. However, nodes may not respond to treatment, or may respond maliciously. These adversarial nodes influence treatment topology by preventing or altering the expected network effect, but may not be known or detectable. We characterize the influence of adversarial nodes and the bias these nodes introduce in average treatment effect estimates. In particular, we derive expressions for the bias induced in average treatment effect using the linear estimator from Gui et al (2015). In addition to theoretical bounds, we empirically demonstrate estimation bias through experiments on synthetically generated networks. We consider both the case in which adversarial nodes are dispersed randomly through the network and the case where adversarial node placement is targeted to the highest degree nodes. Our work demonstrates that peer influence makes causal estimates on networks susceptible to the actions of adversaries, and specific network structures are particularly vulnerable to to adversarial responses.
Link this pageWould you like to put a link to this lecture on your homepage?
Go ahead! Copy the HTML snippet !