Introduction
In the domain of security games, the traditional focus has been on the tactical deployment of a fixed set of resources—answering the question: "Given these $N$ guards, where should they be positioned to minimize risk?" However, real-world security operations are rarely so constrained. Security managers must first decide which resources to acquire or which specialized teams to form within a limited budget before they can even consider deployment strategies.
Sara Marie McCarthy’s doctoral research addresses this gap by introducing the Simultaneous Optimization of Resource Teams and Tactics (SORT) problem. This framework integrates high-level strategic investment with low-level tactical execution, providing a scalable approach to protecting critical infrastructure, natural resources, and digital networks under conditions of significant uncertainty.
The SORT Problem: Strategic and Tactical Integration
The SORT problem is computationally "wicked" due to its two-tier nature. At the Strategic Planning level, defenders must select from a combinatorially large number of heterogeneous resource combinations (e.g., drones, sensors, human patrols). At the Tactical Planning level, the effectiveness of any chosen team must be evaluated by solving a complex deployment problem.
Because the utility of a resource is not intrinsic but depends entirely on how it is deployed, these two stages cannot be solved in isolation. Furthermore, real-world variables—such as noisy sensor data or unpredictable arrival times at checkpoints—ensure that static plans often fail during execution. McCarthy’s research provides three primary algorithmic contributions to navigate these challenges.
Algorithmic Contributions
FORTIFY: Scaling Team Formation
For scenarios like forest protection, where the pool of potential resources is vast, McCarthy developed FORTIFY. This scalable branch-and-bound algorithm utilizes hierarchical abstractions of the security game. By calculating upper and lower bounds on a team's potential utility through a simplified version of the tactical game, FORTIFY can prune the search space effectively. This allows the system to identify optimal team configurations without the need to evaluate every possible combination, making the combinatorial explosion manageable.
Virtually Distributed POMDP (VD-POMDP)
In cybersecurity and network defense, the challenge is often "active sensing"—deciding where to look for threats in a massive, noisy environment. The Virtually Distributed POMDP (VD-POMDP) model decomposes a global decision problem into smaller, localized sub-POMDPs with sparse interactions. A fast Mixed-Integer Linear Program (MILP) is then employed to resolve these interactions at execution time. This decomposition allows for real-time, adaptive responses to security alerts that would otherwise be mathematically unsolvable at scale.
Robust Threat Screening Games (RTSG)
Infrastructure security, such as airport screening, faces the challenge of temporal uncertainty. If a flight is delayed, a "perfect" schedule based on predicted arrival times becomes useless. The Robust Threat Screening Games (RTSG) framework utilizes compact linear decision rules and robust reformulations. This ensures that screening strategies remain feasible and effective even when real-world realizations deviate from the expected schedule, prioritizing strategies that are robust across a range of possible scenarios.
Empirical Evaluation and Real-World Impact
The research is grounded in diverse, real-world datasets and simulations:
- Green Security: Using network and resource data from partners in Madagascar, the research demonstrated how FORTIFY can optimize patrols to prevent illegal logging.
- Cybersecurity: The VD-POMDP model was validated on a real network testbed, where it successfully identified actual cyber-attacks while maintaining high computational efficiency.
- Infrastructure Security: Simulations of airport-style checkpoints showed that the RTSG approach significantly outperformed standard methods. While traditional methods often produced strategies that were physically impossible to implement during peak traffic shifts, the robust approach maintained feasibility and high detection rates.
Key Insights
McCarthy’s work highlights several critical shifts in security game theory:
- Interdependence: You cannot pick the best team without knowing their optimal deployment; strategy and tactics are two sides of the same coin.
- Abstraction for Speed: Simplified models are not just for ease of use; they are essential "speed multipliers" that allow algorithms to rule out suboptimal configurations quickly.
- Robustness over Optimality: In dynamic environments, a strategy that is "good enough" across many scenarios is far more valuable than a strategy that is "perfect" for only one.
Conclusion and Future Directions
This research moves security games from theoretical abstractions toward practical, deployable systems. By accounting for both the cost of resources and the uncertainty of the environment, McCarthy provides a roadmap for more resilient defense strategies.
Future work aims to expand these models into AI for Social Good, applying green security principles to broader social welfare domains. Additionally, the author suggests moving toward multi-stage strategic planning, where investment decisions are made sequentially over years, allowing defenders to adapt their resource pool as new information and threats emerge.
Summary updated