The University of Sydney has launched the Centre for Translational Data Science, which will use data for ‘translation’ to solve real world problems.
This will include crime prevention, reducing youth unemployment, advancing medical treatment and industry efficiency.
The centre comprises academics from all six of the university’s faculties, including the Faculty of Engineering and Information Technologies, the Business School and the Sydney Medical School.
One aim of the centre is to try and develop methods to make new discoveries using data.
“We feel [that we are] problem solvers rather than solution providers necessarily. We go out, we talk to people about their problems and we essentially work with them to try and solve them,” said Professor Hugh Durrant-Whyte (pictured above).
“As opposed to sitting in a group and coming up with new mathematical algorithms and then going out, looking for a problem to solve with them.”
One example of the work the centre is doing is around predicting metal in archaeological deposits in Mesopotamia.
“That data exists through, for example, chemical compositions of the tools people were using at the time and some understanding of where that came [from] in terms of geology,” Durrant-Whyte said.
“The data science is really to then predict models for how trade developed 5,000 years ago and how all those different components are actually linked together.”
Small data problems
Although recent talk around data has focused on big data, Durrant-Whyte said that doesn’t pose a problem. Instead, the challenge is around small data.
“Small data problems are much harder than big data problems because with big data problems you almost always have enough information to come up with an answer,” he said.
“You really need to understand all possible solutions to the problem and the limited amount that you can say given the data you have.”
This is particularly important in fields such as health, where several answers could exist for a problem, but the only way to make a scientific discovery is to test the data, according to Durrant-Whyte.
Another challenge is refining what question or problem needs to be answered.
“Too often we go into a project and the biggest problem we have is getting them to understand what the question is that they wish to answer,” Durrant-Whyte said.
“Not whether they can get a piece of information, but what are they actually trying to do if they had all that information, in terms of making a difference, making a decision and so on?”
Fields that involve people can be particularly challenging, such as mental health.
Durrant-Whyte said one reason is because why and how people suffer from mental illnesses can be extremely complex and very little data exists.
“You really have to work hard to build proper models and to articulate and explain all the uncertainties that are involved,” he said.
“Those are really hard areas to get involved in. Humans particularly because samples are very different and in humans there are complex things that really go up to build up that data set.”
For research around mental health, Durrant-Whyte said data can come from an individual with a condition, such as their life history and how they came to be in the situation they’re in. This could include aspects such as whether they were abused as a child.
“All we know in terms of the data is where they have ended up,” Durrant-Whyte said. “Now what we need to understand is as a consequence of that individual, what’s the right treatment for them? … The space of possible things you need to do is huge.
“As we head towards an age of hopefully personalised health care, this is exactly the kind of data strategy that people will need to adopt.”
The centre is also doing work around domestic violence using crime data that has been provided by the police and other sources, with the aim of helping the police understand why domestic violence happens and how to reduce it.
But although there are reported cases the team can rely on, the data is not indicative of the complete picture, which means estimating unobserved and unmeasured aspects.
It also involves questioning what aspects influence domestic violence, such as location, economic factors and cultural factors to help prevent crime instead of just dealing with the after effects.
“The spacial temporal model is an important part of that because it lays out in some sense how these variables are linked together, and in some sense again, the way that you might want interventions to actually occur,” Durrant-Whyte said.
“In criminology, [spacial temporal models] might be relatively rare, but we do use them in other areas. For example, understanding people in transport or understanding ecology and those sorts of things. We’re really taking statistical techniques and applying them to this problem.”
The centre has already garnered results in mental health, but Durrant-Whyte said he’s waiting to see what cultural change takes place.
“I think that takes time. It’s not just technically solving it, which is really often the easy bit, it’s the cultural shift that needs to happen with people accepting that kind of information,” he said.