Bayesian Network Meta-Model for Causal and Decision Analysis of Combat Simulation

Location: Adelaide, SA

Duration: 4-6 months

Please note: Due to funding requirements, students must have Australian or New Zealand Citizenship or Permanent Residency to apply. Any applicants not meeting this requirement will automatically be deemed ineligible for this project.

Project Background

DST Group use combat simulation as an analysis tool to explore the potential impact of modifying technologies, tactics, concepts, and force structures in the context of military operations. Combat simulation for analysis requires modelling physical characteristics of opposing forces and terrain, behaviours that represent military tactics, techniques, and procedures for manoeuvre and engagement against an active enemy. In a combat simulation model, inputs are often described as the factors affecting the combat effectiveness of the system, i.e., system settings and configurations as well as friendly and opposing forces and operating environments. Simulation outputs, on the other hand, are artificial observations of the system produced by the simulation model. In a simulation study, simulations are performed with alternative values of the inputs and the observed outputs are recorded.

The data are then used to draw inferences concerning the operating characteristics of the system. A combat simulation model, although simpler than real-world war-fighting, can still be complex. The repetition of simulations may be time consuming and the sheer size of simulation data sets can make them unwieldy. To avoid this inconvenience, simulation meta-models are used to represent the dependence between simulation inputs and outputs. The most commonly used meta-models are input-output mappings that project the values of simulation inputs to the expected values of outputs. They include: regression models, Kriging, Neural networks and Bayesian Networks (BN), etc.

In this research proposal, we propose to use BN as probabilistic multiple input and multiple output simulation meta-models. A BN meta-model is a representation for the joint probability distribution of random variables associated with simulation inputs and outputs and is used to calculate marginal and conditional probability distributions as well as expected values and other descriptive statistics related to the inputs and the outputs. That is, the complete probability distributions are modelled – not just expected values as is the case with the existing meta-models. The BN meta-model enables various what-if analyses that are used for studying the marginal probability distributions of the outputs, the input uncertainty, and the dependence between the inputs and the outputs. Additionally, it is used to examine the dependence between the outputs and perform inverse reasoning. As far as we know, such analyses are beyond the scope of the existing MIMO simulation meta-models. The BNs allow effective what-if analyses which could be time consuming if conducted based on raw simulation data. Overall, the BNs offer a flexible approach to metamodeling of combat simulation. An efficient and robust BN meta-model would enhance the analysis of combat simulations including what-if, inverse, causal and decision analysis, thereby increasing their usefulness to a wider scope of problems, as well as reduce the computational effort. This would result in enhancements to DST Group’s capacity and capability to use combat simulation as an analysis tool.

DST Group is seeking researchers with expertise in computational mathematics, statistics or computer science. This problem will require the researcher to discover and apply a combination of conventional and new concepts to deliver a successful solution.

The researcher will be tasked with exploring generic examples of such solutions which can be consistently implemented across a range of simulations. The actual implementation and testing of such solutions in combat simulation will be executed in collaboration with technical simulation SME’s.

Research to be Conducted

The objective of this project is to review and research a suite of BN learning algorithms to construct a BN meta-model from combat simulation data. The researcher is expected to conduct the following study:

  1. Review BN learning algorithms
  2. Structure BN meta-model from an exemplar combat simulation data
  3. Validation of BN meta-model based on different learning algorithms
  4. Analyse the impact of discretisation on validation
  5. Provide what-if, causal, sensitivity and decision analysis of combat simulation
  6. Exploring the feasibility of Dynamic BN (DBN) modelling option for analysing the dependency between combat simulation events (optional)

Skills Required

If you’re a PhD student and meet some or all the below we want to hear from you. We strongly encourage women, indigenous and disadvantaged candidates to apply:

  • Expertise in Computational Mathematics, Statistics, or Computer Science
  • Familiarity with programming or scripting
  • Proficiency in at least one programming or scripting language
  • Expertise in Machine Learning and Artificial Intelligence
  • Military domain knowledge or experience

Expected Outcomes

The exact deliverables will be negotiated based on the researcher’s skillset.
Essential outcomes of this project include:

  1. A literature review of BN learning algorithms applicable to structuring BN meta-model from combat simulation data, delivered as a research paper or report.
  2. Structuring and validating BN meta-model including the impact of discretisation
  3. Provide what-if, causal, sensitivity and decision analysis of combat simulation study

Desirable outcomes include:

  1. A study on the feasibility of applying DBN to the analysis of combat simulation

Additional Details

The intern will receive $3,000 per month of the internship, usually in the form of stipend payments.

It is expected that the intern will primarily undertake this research project during regular business hours, spending at least 80% of their time on-site with the industry partner.  The intern will be expected to maintain contact with their academic mentor throughout the internship either through face-to-face or phone meetings as appropriate.

The intern and their academic mentor will have the opportunity to negotiate the project’s scope, milestones and timeline during the project planning stage.

Applications Close

29 January 2020


APR – 1117