Skip to main content
A conceptual image of light representing data fusion
Research_

Data fusion

Developing techniques to fuel meaningful research
Data fusion aims to develop the mathematical, statistical and algorithmic structures needed to build models of complex phenomena. Integral to this process is ensuring those models are predictive, testable and data centric.

Building predictive and testable data-centric models of complex phenomena in the physical, life and social sciences relies on data fusion – that is, developing fundamental mathematical, statistical and algorithmic infrastructure. Crucial to the process of discovery is the ability to make meaningful inference from observations that are varied in type, volume and precision. The theme will focus on the development of novel probabilistic models and methods to estimate them. These models provide a principled way to fuse information from multiple sources; information from observations, and information from domain specific theories, expert opinion and prior studies.

Techniques

  • Reproducible research
  • Bayesian non-parametrics
  • Causal models
  • Bayesian networks
  • Deep learning
  • Statistical modelling
  • Dimension reduction
  • Missingness

Projects

  • Machine learned defence to adversarial RF systems via system identification and strategy optimisation
  • Origin Energy – modelling subsurface geology
  • Probabilistic datacubes for integral field spectroscopy
  • Localising geological formation boundaries via Bayesian joint inversion
  • Detrending radio transient light curves from the Murchison wide-field array
  • Markov chain Monte Carlo on function spaces
  • Bayesian risk prediction instruments
  • Inferring explosion mechanisms for type Ia supernovae
  • Space-time and demographics crime prediction
  • Academic capability mapping
  • Bias removal in MRI datasets
  • Bayesian optimisation for GLM applied to police patrolling
  • Modelling of crime using features from the environment
  • Modelling of crime using deep learning
  • Visualisation of spatial temporal density functions
  • A quantitative analysis for understanding the relationship between electronic gaming machines, problem gambling, socio-demographic variables and crime
  • Bayesian deep learning for uncertain environments
  • Irreproducibility; nothing is more predictable
  • Coevolutionary multi-task learning for feature-based modular pattern classification
  • Bayesian neural learning via Langevin dynamics for chaotic time series prediction
  • Spectral estimation for non-stationary multivariate timeseries
  • Applying machine learning to criminology: semi-parametric spatial-demographic Bayesian regression
  • A comparison of two models of dental care for Aboriginal communities in New South Wales
  • Bayesian geometric models for nutrition
  • Bayesian hierarchical model for the causes and consequences of a reduction in gestation length