Risk-Sensitive (CVaR) Swarm Optimization for Resilient Resource Deployment in Adversarial Distributed Networks

• Karim Mostafa

  Department of Computer Science, Delta Valley University, 14 Al-Nour Extension, Mansoura 35514, Egypt

  Author

• Youssef Elsherif

  Department of Software Engineering, Nile Crest Institute of Technology, 27 El-Madina Al-Monawara Road, Giza 12556, Egypt

  Author

• Nourhan Abdelkader

  Department of Information Systems, Red Sea Applied Science College, 8 Port Boulevard, Hurghada 84517, Egypt

  Author

Distributed networks that allocate mobile or stationary resources are often exposed to structured disturbances and malicious attacks. Routing, sensing, and protection functions are executed by swarms of agents that have limited local information and must coordinate through sparse communication patterns. In this setting, purely expectation-based optimization can lead to fragile operation because rare but severe disruptions are not explicitly controlled. Risk-sensitive formulations provide an alternative by incorporating tail events into the design objective. Among these, the conditional value at risk offers a convex representation of loss in the worst upper tail of the distribution. Swarm optimization methods provide a complementary mechanism for exploring high dimensional deployment spaces through the collective motion of agents driven by simple interaction rules. This paper considers the combination of a conditional value at risk objective with swarm based coordination for resource deployment in adversarial distributed networks. The approach models adversarial actions as random but structured perturbations of network components and computes deployment policies that hedge against the induced tail losses. The resulting formulation is expressed as a linear program over a scenario based representation and is approximated in a fully distributed manner by a population of agents. The paper develops the modeling, algorithmic, and analytical aspects of the framework, and discusses numerical experiments that illustrate the behavior of the method under different adversarial intensities, network topologies, and communication constraints for the swarm.

• PDF

Copyright (c) 2018 authors

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.