This AI researcher and entrepreneur is passing on her commercialisation expertise to the next generation.
The way Professor Svetha Venkatesh describes it, machine learning is simply the search for patterns. Like sorting apples from bananas.
“Humans do this unconsciously, right? We say that anything which is red and round is an apple, anything yellow and kind-of elongated, we’re going to call a banana,” she told The Brilliant. “That’s a pattern. Machines find much more complex patterns automatically in data, so that they can categorise, describe and represent the different types of objects in the world.”
From the simplicity of the fruit bowl to the complexity of the modern world: Venkatesh’s machine learning discoveries can triage patients at risk of suicide and help parents teach their autistic children. They’re creating new materials for manufacturing and, in the aftermath of the tragic London Bombings, led to a breakthrough in surveillance technology to make cities safer.
Named as one of the top 15 women in the world in AI, Venkatesh is Alfred Deakin Professor at Deakin University in Victoria and co-director of its Applied Artificial Intelligence Institute. In 2021 she was named a Fellow of the Australian Academy of Science for her ground-breaking discoveries.
Apart from an extraordinary breadth of research experience, she also knows how to commercialise research and create impact. To speak with her is to appreciate the scope of problems that machine learning can crack when experts collaborate.
“Talking to people in other disciplines sometimes gives us a new machine learning problem that we need to solve,” says Venkatesh. “I don’t think of working on these problems as an application of machine learning. I think of it as a co-journey that people from a different field and I are going on. They influence me and I, hopefully, can influence them.”
An eye for answers
Born into a family of doctors and engineers in India, Venkatesh studied electronic engineering and moved into the fledgling computer industry in the early 1980s before shifting to Perth. With a young son to care for, she switched to working in academia for its greater flexibility. And she focused on machine learning and AI, a field bursting with potential thanks to increasingly powerful computers and a mass of new data, like the footage captured by security cameras.
Venkatesh and colleagues at Curtin University were working with the Perth-based DTI Group, which had installed security cameras in buses in London, when suicide bombers attacked the transport system in 2005, detonating devices in trains and a bus, killing 52 and injuring 700. The city’s confidence was rocked by terrorism enacted by young men with backpacks.
Crime investigators wanted to know if one of the bombers had gotten off the bus before the device detonated. With cameras inside and outside the bus, it should have been an easy question to answer, yet no-one could extract detail from the footage.
“We started from that point,” Venkatesh explains. “‘Why is all this footage so unusable? What are the problems we need to solve in this space of wide-area surveillance?’”
The result of her team’s research was Virtual Observer, a commercial software system that enables access to footage from the many mobile cameras – on security vehicles and buses, for instance – using GPS information to pinpoint each one’s position at any given time. In the absence of CCTV footage, Virtual Observer identifies and, potentially, supplies the missing detail, making cities safer.
Security cameras featured in another breakthrough. Still working with DTI as well as the Perth Transport Authority, she visited a security centre at a major train station and chatted with the security guards.
“There were like 8000 or 10,000 cameras on the system, and only had five screens to monitor them! How do they know which camera to look at? One guard said, ‘I know from experience,’ but I knew you couldn’t really do that.
“So that’s the problem we set out to solve. First, you learn what is ‘normal’ for each camera, and then detect what is ‘abnormal’ for each camera,” she says. “The system only alerts the operator when something is abnormal; the operator doesn’t choose, the system chooses for them. And that is the basis of the startup icetana that we commercialised.” More than a decade later, icetana is a successful ASX-listed company that deploys its products worldwide.
Venkatesh’s experience as a researcher with startups like icetana inspired her to create SPARK Deakin in about 2017: the program aimed to develop commercialisation skills in next-gen science entrepreneurs.
Skill number one, according to Venkatesh, is to understand that the CEO is key. “One of the mistakes people make is to think that the startup is the product. The startup is not the product; the startup is to find the business model to make that product work,” she says. “It requires completely different skills from the kind of skills I have. We need to work as a team, which is what I did with Matt [Matthew Macfarlane, founding and current CEO of icetana].”
Isn’t it tough to hand over your precious discovery to an ‘expert’ who isn’t you? “I’m a researcher,” she shrugs. “I can always move on to my next problem. I guess if you were working on the same thing your whole life, it might be very hard to do. But for me, it’s not so much of a big deal.
“I didn’t ever really think about it from a commercial perspective. I always wanted to solve some problem for which there is no solution, with machine learning.”
Using AI to overcome human struggles
In Perth, Venkatesh saw first-hand the struggles of parents with autistic children wanting to access early intervention services. Waitlists for professional intervention were long and frustrating, so she used machine learning to create TOBY (Therapy Outcomes By You) – software that gives parents a framework for helping their children work on language, social communication and other skills. Then available on the App Store, it was accessed by thousands.
More recently, she has focused her research energy in applying machine learning to one of the most urgent health issues of our time: mental health. Her work with a Geelong-based healthcare provider illustrates machine learning’s powerful impact.
The organisation asked Venkatesh’s Deakin team to identify which patients were at higher risk of suicide. But the predictive patterns the team needed simply weren’t evident in data generated by doctor–patient communication.
Instead, surprisingly, it was billing data that provided the basis of a triage system to classify clients as being at low, medium or high risk of suicide.
Venkatesh’s team is currently working with the Black Dog Institute, using machine learning to accelerate trials to test the effectiveness of mental health treatments, and is also collaborating with material scientists on ways to accelerate the design of new materials for manufacturing. “I’m very drawn to any new science,” she says.
In 2019, Venkatesh made a global list of the top 15 women in AI, created for a research paper on gender diversity in the AI workforce. The lack of women is still a problem, she concedes; as head of Computer Science at Curtain University she worked on ways to increase gender diversity but “never quite cracked that. A lot needs to be done at the school level,” she says. Venkatesh welcomes the advent of programs like Girls Who Code but sees a long road ahead before there is more of a gender balance in the AI space.
In the meantime, here’s her sales pitch to young women considering a career in AI: “If you like mathematics and building systems, then this is a very interesting field that presents you with opportunities to innovate.”
Article by Michelle Fincke
Comments are closed.