Cognitive Data analytics with Horticulture Innovation Australia and The Yield
I’m a researcher on QUT-Industry grant with Horticulture Innovation Australia project VG15054: Data analytics and app technology to guide on farm irrigation and pest management 2016-2018. My role is to contribute expertise in cognitive data analytics to maximise the usability and adoption of apps created by agricultural start up The Yield.
Agri-Intelligence in Cotton Production Systems
I’m a research fellow with the Institute of Future Environments on Cotton Research Development Corporation (CRDC) Grant: Agri-Intelligence in Cotton Production Systems (2016-2018) developing a decision making framework for high risk farming decisions under uncertainty.
This project (2016-) aims to build a Bayesian dashboard for data engagement, analysis and insight. More and more businesses use visual analytics dashboards to represent their data for better decisions. However, typical analytics dashboards don’t actually improve decisions much because their data doesn’t enhance the cognitive firepower of users. Before leaders can analyse data, they need to understand the function of the data in their own reasoning processes. We propose a new sort of dashboard—a Bayesian dashboard.
Bayesian rationality dictates that leaders need to generate likely hypotheses about their business before they even consider visualising data . If leaders specify a robust set of likely hypotheses and set prior probabilities, then a Bayesian dashboard would get populated with data visualisations in the service of these hypotheses. A Bayesian dashboard ought to represent data as evidence to activate human reasoning. Visualised data becomes evidence for or against multiple hypotheses. Users interact with the dashboard by making bets on which hypothesis they think is most likely to be true, similar to making bets on a horse race. Dashboard predictions could be linked between users.
While there are Bayesian dashboards already in existence, they are designed to feed Bayesian networks (i.e. machine learning algorithms) or to create Bayesian decision support systems, such as personalised drug treatments . I believe this project would be the first attempt to make a Bayesian dashboard for human reasoning purposes. I propose that such a dashboard would generate greater user engagement with data, data comprehension, analysis and insight. There would be a feed-forward loop of user data that leaders could use to generate new hypotheses for further consideration in the dashboard.
A Bayesian dashboard will promote and assist data analysis for increased situational awareness and new insights leading to better resilience.
 Montibeller, G., & Winterfeldt, D. (2015). Cognitive and Motivational Biases in Decision and Risk Analysis. Risk Analysis, 35(7), 1230-1251. doi:10.1111/risa.12360
 Reddy, Vikas, Farr, Anna Charisse, Wu, Paul P., Mengersen, Kerrie, & Yarlagadda, Prasad K.D.V. (2014) An intuitive dashboard for Bayesian network inference. In Journal of Physics: Conference Series, Institute of Physics Publishing Ltd., 012023. Retrieved from: http://eprints.qut.edu.au/63346/
 Mould, D., D’Haens, G. and Upton, R. (2016), Clinical Decision Support Tools: The Evolution of a Revolution. Clinical Pharmacology & Therapeutics. doi: 10.1002/cpt.334
Using expert intuitions with Bayesian models for increased productivity
Many industries have expertise built up over decades, yet no strategy or processes on how to capitalise on this expertise to improve productivity. Expertise is important because pure mathematical (e.g. Bayesian) models of decision-making struggle to fix the prior probabilities of hypotheses Hm to Hn important to evaluate increasingly big data. My research investigates how experts, via their years of experience, pitted against theoretical knowledge, have insight into large data sets to help set prior probabilities where abstract models fail. More importantly, the intuitions of experts are reliable in a way that the intuitions of novices are not. This expertise can be captured if we set up a process by which experts are invited to judge the likelihood of particular hypotheses, and then automated processes can be implemented to evaluate evidence algorithmically.
My hope for research over the next three years is to investigate how Bayesian models of decision-making can be improved with systematic, higher order, reflective human expertise to solve real world issues in big data utilization and analysis . I hypothesize that experts, via their years of experience and honed theoretical knowledge, have insight into large data sets to
- help set prior probabilities
- decide what data is relevant
- offer reliable intuitions
- prioritise evidence (e.g. weight causal propositions more highly than particulars)
- conceive of likely hypotheses and
- judge the coherence of partitions.
In the course of the research process I envisage a) creating more effective autonomous Bayesian models b) systematically mapping expertise ontologies and c) gaining insight on normative human/machine decision-making.