Leveraging Bayesian inference and machine learning: Dynamic Bayesian Networks for managing resilience

Study level


Master of Philosophy


Vacation research experience scheme

Topic status

We're looking for students to study this topic.


Distinguished Professor Kerrie Mengersen
Australian Laureate Fellow
Division / Faculty
Science and Engineering Faculty
Dr Paul Wu
Senior Lecturer in Statistical Data Science
Division / Faculty
Science and Engineering Faculty


Dynamic Bayesian Networks (DBNs) provide a way to represent complex systems and their evolution over time by capturing cumulative local interactions between variables to infer global effects. They feature great flexibility and widespread applicability as demonstrated by applications in ecology, medicine, genetics, logistics and many more.

Coming from machine learning and statistics, they are a generalised form of Hidden Markov Models (HMMs) and Kalman Filters. Here at QUT, exciting new developments of DBNs for systems that change over time led to recent findings to manage environmental impacts using timing as reported in Nature Communications.

Research activities

In this project, the student will explore the intersection of machine learning (DBNs) and Bayesian inference. Bayesian inference provides a systematic and rigorous approach for prediction and estimation of real world problems under uncertainty. Measurement error and uncertainty in conditional probabilities present a challenge for DBNs.

Incorporating Bayesian paradigms in DBN inference could leverage the benefits of both statistics and machine learning to better estimate DBN parameters for complex systems and provide more robust estimates of uncertainty in model predictions. Opportunities to apply this research include data and collaborations with IFREMER (Brest, France) for seagrass resilience management, and Swimming Australia and Queensland Academy of Sport for athlete resilience.


The outcomes of this research depend on the specific topic chosen by the student under the umbrella of modelling and study of resilience for complex systems.

Potential methodological outcomes include:

  • Novel, computationally efficient methods for Dynamic Bayesian Network inference for large spatial and temporal scales, such as the study of climate change effects on seagrass
  • Novel Bayesian approaches for learning and inference for Dynamic Bayesian Networks taking into account measurement uncertainty, such as uncertainty in athlete stress and recovery studies,
  • Novel insights to manage impacts on critical marine ecosystems such as seagrass using these complex time-series models,
  • Novel insights to manage for maximising performance and minimising injury for swimmers.

Skills and experience

The project details could be adapted to a student's mathematical/statistical and applied interests. It is likely to require the development and software implementation of sampling methods such as MCMC, inference and evaluation of conditional probabilistic networks, and practical application of the methods to real-world problems with industry partners in ecology and/or sports. Optionally, students could focus on computationally efficient inference algorithms / samplers. Or, students could focus on model development and application, as suits the students particular interest.


You may be able to apply for a research scholarship in our annual scholarship round.

Annual scholarship round



Contact the supervisor for more information.