A Strategic Framework for Improving Resilience in AI and General Adversarial Cyber-Physical Networks
Name
pan-luisap-meng-eecs-2026-thesis.pdf
Size
64.36 MB
Format
Adobe PDF
Checksum (MD5)
543ea78c8ec2dd6d1414db6330c3647f
Author(s)
Pan, Luisa C.
Advisor(s)
Pal, Ranjan
Siegel, Michael D.
Date Issued
February 2026
Publisher
Massachusetts Institute of Technology
Abstract
This thesis introduces a strategic framework improving cyber-physical network resilience against evolving threats, including LLM-generated attacks. Motivated by critical infrastructure vulnerability to cascading failures, the study modeled cyber defense as a time-constrained resource management problem embedded within a network flow optimization task. MixedInteger Programming (MIP) optimized resilience—defined as normalized cumulative functionality restoration—evaluated via a large-scale Monte Carlo simulation framework across IEEE 30-bus and 118-bus systems. We analyzed how efficient traditional graph-theoretic heuristics (Degree, Betweenness, Spectral, Mincut) compared to newer strategies generated by Large Language Models (LLMs). Results revealed traditional heuristics were limited, prioritizing individual edge criticality rather than holistic edge sets that disrupt a network. However, strategies causing widespread disruption (such as AI and random) proved superior by focusing on difficult-to-address system-wide targets. Findings also highlighted the superiority of strategic, weighted defense resource allocation over equal distribution. Moreover, recovery time constraints outweigh organizational business types in determining resilience outcomes, while larger network scales stabilize degradation through increased redundancy.
Keywords: cyber resilience, edge centrality, Artificial Intelligence, Monte Carlo simulations, Large Language Models
MIT Department
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
Terms of Use
In Copyright - Educational Use Permitted
Copyright retained by author(s)
Persistent DSpace Link
