Modeling and simulation are critical capabilities for designing vehicles and infrastructure, forecasting markets and weather, investigating physical processes, and developing and improving strategies for incident response, battle management, and disaster recovery. The more complex, extensive, and interdependent the system, product, environment, or event, the longer it takes a simulator to yield useful results. For challenging problems, such as multi-domain warfighter planning, simulation time is a bottleneck even on leading edge high-performance computing platforms.
Peraton Labs harnesses artificial intelligence and analytics to achieve multi-fold speedups in simulation studies. Our novel techniques use machine learning (ML) and leverage data, particularly intermediate and final results of past simulations, to turbocharge simulation by:
Consider the challenge of determining effective operational strategies for responding to a significant cyberattack on U.S. critical infrastructure, such as malicious contamination of a city’s water supply[1], a transportation shutdown in a major metropolitan area[2], or an extensive, nation-wide disruption of internet service[3]. The simulation problem space incorporates assumptions regarding the defensive techniques and tools in place, the types and amounts of resources for responding, and the possible countermeasures by the attacker. Each simulation experiment defines one scenario determined, for example, by the location, status, and availability of infrastructure, mobile emergency equipment, trained staff, etc. Using models for resource demand, attacker behavior, and responder actions, multiple discrete event simulation runs are conducted to identify outcomes, their likelihoods, and best course of action for that scenario. The simulation study investigates the problem space to answer key questions:
Our techniques are designed to achieve confident answers to these questions quicker and at less cost. They leverage the simulation data, including intermediate and partial results, to enable multi-fold reductions in the simulation effort for complex operational and planning environments.
Contact us at [email protected] to learn more.
[1] Florida water treatment plant cyberattack symptom of larger problem (techgenix.com).
[2] Navigating the threat of cyber attacks on the transport sector | TechRadar
[3] Rogers network resuming after major outage hits millions of Canadians | Reuters
Peraton Labs harnesses artificial intelligence and analytics to achieve multi-fold speedups in simulation studies. Our novel techniques use machine learning (ML) and leverage data, particularly intermediate and final results of past simulations, to turbocharge simulation by:
- Reducing the number of simulation experiments (or scenarios)
- Reducing the number of simulation runs per experiment
- Reducing the execution time of a simulation run
Consider the challenge of determining effective operational strategies for responding to a significant cyberattack on U.S. critical infrastructure, such as malicious contamination of a city’s water supply[1], a transportation shutdown in a major metropolitan area[2], or an extensive, nation-wide disruption of internet service[3]. The simulation problem space incorporates assumptions regarding the defensive techniques and tools in place, the types and amounts of resources for responding, and the possible countermeasures by the attacker. Each simulation experiment defines one scenario determined, for example, by the location, status, and availability of infrastructure, mobile emergency equipment, trained staff, etc. Using models for resource demand, attacker behavior, and responder actions, multiple discrete event simulation runs are conducted to identify outcomes, their likelihoods, and best course of action for that scenario. The simulation study investigates the problem space to answer key questions:
- How do current defense mechanisms perform? Can they be defeated by a canny adversary?
- What resources, deployment strategies, and command structures are most effective in mitigating the attack? What are the optimal courses of action?
- What is the impact on defensive capability, courses of action, and attacker countermeasures of extreme conditions or unexpected occurrences (e.g., a severe weather event)?
Our techniques are designed to achieve confident answers to these questions quicker and at less cost. They leverage the simulation data, including intermediate and partial results, to enable multi-fold reductions in the simulation effort for complex operational and planning environments.
- Our experiment optimizer provides analytic estimates of the effect of inputs on the outcomes. This refines a large simulation problem space into a smaller space of interest, reducing the number of experiments.
- Our ML-based predictor provides rapid what-if analysis to estimate performance and error bounds for unexplored regions of the input space. This avoids unnecessary experiments and maximizes the value of the simulation data.
- Our hyperspace explorer provides intelligent search over the model hyperspace and recommends design points with a high probability of meeting the performance outcomes for the scenario. This reduces the number of simulation runs required to achieve confidence for an experiment.
- Our fragment similarity estimator reduces the execution time of a simulation run by identifying repeating fragments in the simulation (e.g., steps to deploy emergency response helicopters in a certain environment). The fragment similarity estimator uses sophisticated reasoning and machine learning to recognize repeats, replace them with an ML-estimate, and speed up execution.
Contact us at [email protected] to learn more.
[1] Florida water treatment plant cyberattack symptom of larger problem (techgenix.com).
[2] Navigating the threat of cyber attacks on the transport sector | TechRadar
[3] Rogers network resuming after major outage hits millions of Canadians | Reuters
