Title:

Modeling and Mitigation of Network Cascades

Abstract:

As our world becomes more connected through physical, technological and socioeconomic networks, it becomes more vulnerable to failures in these networks. These failures, which often rapidly diffuse across the network, have dealt serious societal damage in the form of epidemics, computer viruses, and misinformation. It is therefore imperative to understand how such "cascading failures" propagate, and how interventions can be designed to mitigate their effects.

In the first part of this talk, I will discuss recent progress in modeling network cascades. A fundamental hurdle in the analysis of cascades is their inherent high-dimensional structure, caused by the complexity of the underlying network as well as the stochasticity of the cascade dynamics. This motivates the use of simpler approximations for understanding the evolution of the cascade. I will touch upon new results along these lines related to mean-field models, mask-wearing, and multi-strain models with mutations.

In the second part of this talk, I will develop algorithms for the real-time localization of network cascades. Specifically, our goal is to identify the cascade source before too many vertices in the network are affected by the cascade. The cascade is assumed to spread according to a Susceptible-Infected process from an unknown source in a network. While the propagation is not directly observable, noisy information about its spread can be gathered through multiple rounds of error-prone diagnostic testing. Using this data model, we devise a novel adaptive procedure inspired by classical multi-hypothesis sequential probability ratio tests (MSPRTs) which provably localizes the cascade before a negligible fraction of the network is affected. In certain cases, our method is optimal, i.e. no other algorithm can localize the cascade using substantially fewer rounds of testing. Based on joint work with Tirza Routtenberg and H. Vincent Poor.

Bio:

Anirudh Sridhar (Ani) is a postdoctoral associate at MIT's Department of Mathematics. Previously, he completed his PhD from Princeton's Department of Electrical and Computer Engineering, where he was advised by H. Vincent Poor and Miklós Rácz. Broadly, Ani's research develops statistical methods for the analysis of networks, with a focus on information-theoretic characterizations. His awards include the Yan Huo *94 Graduate Fellowship in Electrical Engineering from Princeton University in 2022 and a Spotlight Presentation at NeurIPS 2021. He was also a finalist for the Informs-APS Best Student Paper Award in 2020.