Machine learning aims to make predictions about novel data based on associations and relationships from past data. This approach rests on the assumption that the past is representative of the future.
We're exploring how statistical machine learning might be used knowing that aspects of the future are likely to be fundamentally different to the past that generated the historical training data.
Our aim is to is to look for fruitful ways to couple simulation and inference so that we can take advantage of:
- past data
- understanding that our future will be different
- the ability to simulate data from an explainable model.
This project is inspired by the situation faced by government and other decision makers in the wake of the COVID-19 pandemic. Aspects of Australia’s future will be very different to the past embodied in various historical data sets. Are there practical ways for us to still use that data in conjunction with beliefs about our systems and possible future scenarios?
You will explore relevant literature and formulation of more focused research questions that can be meaningfully addressed within the available time and resources. You'll have the opportunity to interact with academics in QUT's Centre for Data Science.
We expect this project will organise our thinking about the potential for statistical machine learning to be used in conjunction with simulation.
One output will be a review of relevant literature as well as a framework and pilot system that can be used to make inferences from historical and simulated data.
Skills and experience
Ideally, you'll have excellent skills in critical and creative thinking, coupled with strong quantitative capabilities. You should be able to work effectively without close supervision but with appropriate engagement and consultation.
This is a project for someone whose research capabilities are relatively mature and who would like to make this topic their own.
Contact the supervisor for more information.