NETWORK ARCHITECTURES FOR COMBATING NOISE AND MALICIOUS ACTIVITY
Abstract
We propose to develop a new science of graphs and their related resilience under the distributed computation of their graph properties in the face of a malicious attack. We focus on the derivation of 1) detection methods that develop distributed methods for detecting vulnerabilities in the topology and detecting compromised agents in the network and 2) actuation methods, or ways to control the network topology to restrict the circulation and impact of compromised data in the network. Our main tools to achieve this will involve i) the investigation of several concepts from network theory including- page rank, leverage scores, machine learning, and leveraging neighborhood observations (opinions of trust) for deriving fast detection methods, and ii) the investigation of several network topologies such as- switching topologies, and hybrid networks which are a combination of local clustered networks and intermittent access to a centralized server or cloud. For the case of hybrid or switched network architectures, we exploit the particular advantages that these architectures have for combatting adversarial agents in the network. In the case of hybrid networks, intermittent opportunities to communication with a cloud server, or other agents in the network with heterogeneous capabilities in terms of information, computational resources, or hardware resources, can be strategically exploited to endow the entire network with more resilience. For the switched network case we consider the benefits of intentionally breaking certain links in order to thwart the spread rate of potentially malicious information in the network. By doing this, we target the ability to bound the impact of an attack on the system. Finally, we study dynamic graphs where we control agent mobility and their consequent neighborhoods (graph edges), and thereby design the network architecture to best repair vulnerabilities and-or enhance its inherent resilience.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 07, 2023
- Source ID
- FA95502210223
Entities
People
- Stephanie Gil
Organizations
- Air Force Office of Scientific Research
- President and Fellows of Harvard College
- United States Air Force