Lead CIs: Kerrie Mengersen
Bayesian methods for modelling and analysis of data are now well established. However, as with many statistical methods, their applicability in the context ‘big data’ is still being explored. This project will consider four questions that relate to ‘scalable Bayes’, that is, Bayesian models and computational methods that scale up to big data problems.
- Bayesian models for data integration: development of a general framework for integrated analysis of data of different type from different sources. This includes observational data, sound and other sensor data, image data, genetic data, streaming data, online and other digital data. The envisaged framework will comprise a hierarchical combination of different parametric and nonparametric approaches.
- Bayesian models for large observational studies: development of approaches to describe and analyse large scale observational studies such as those found online. These approaches will include, among other things, new experimental designs and the use of propensity scores.
- Computational algorithms: pursuit of computational methods for Bayesian analysis that scale up to high dimensions. These will include Approximate Bayesian Computation and sequential Monte Carlo.
- Visualisation of Bayesian model outputs: new ways to visualise the results of Bayesian analyses.
The methods developed as part of 1-5 above will be applied to the three Collaborative Domains:
- Healthy People: this will interface with the Systems Medicine project led by Kevin Burrage.
- Sustainable Environments: in collaboration with PO AIMS, the aim will be to develop multiscale mathematical and statistical models and fast computational algorithms that utilise survey data, streaming data obtained from environmental sensors, remotely sensed data at different spatial and temporal scales to better understand environmental impacts and trends in biodiversity on the Great Barrier Reef.
- Prosperous Societies: this will be undertaken in collaboration with colleagues at Google and e-Bay Research Labs, and ABS.