In her own words
I was very STEM-bent from a young age and loved looking at patterns. When I was 12, I took an extra-curricular course to learn coding as I was fascinated by computers. I was the only girl in the class. But I didn’t focus on that. I had a thick skin and was driven by my curiosity. In fact, it is curiosity which continues to drive me as a researcher. I am drawn to solving new problems and that excites me. I decided to go into computer science and my interest in understanding patterns led me to work initially in data mining and pattern recognition.
My research is at the intersection of AI, ubiquitous computing, behaviour modelling, and machine learning using multimodal sensor data. This work includes capturing mobility behaviour and supply and demand patterns and trends. The challenge is to extract intelligence from all the big data coming out of multimodal sensors to provide insights for decision makers around transport planning and investment.
Historically transport planning relied on creating stimulations, involving a lot of assumptions to generate scenarios and outcomes. The issue is many of our assumptions are based on what we know has worked in the past. But the world has changed post pandemic, and people have not returned to their offices five days a week. So, the model you built in 2018-19 won’t necessarily work now. Furthermore, climate change is increasingly impacting our transport networks, with rising numbers of floods and heatwaves, and this cannot be modelled just by using assumptions.
It is only recently that transport planning has become data driven and we can better understand how our cities behave under different circumstances and conditions.
With Generative AI, we can harness both data-driven modelling using past observations, as well as the powerful knowledge from pretrained models, such as Large Language Models (LLMs), to be able to generate more accurate forecast and improve the simulation of future scenarios. We are one of the first groups in the world that have leveraged generative AI and LLMs for mobility modelling. Our publication and patent (pending) on natural language and prompt-based interaction for human mobility forecasting was published in early 2022, many months before the release of ChatGPT.
In my previous role at RMIT, I worked with the City of Melbourne on their pedestrian traffic flow data and built a model that can predict the number of people that will be in different locations in the Melbourne CBD in the next 14 days at each hour to inform crowd management and planning. I also developed parking analytics for the Mornington Peninsula Shire, with our sensors and monitoring systems deployed at the start of the COVID lockdown, in early 2020, when people’s behaviours dramatically changed, as most people were home due to the lockdown. We built a predictive analytics system for parking availability, and this model should have worked even after the lockdown was lifted, but there was not enough training data of people’s travel and parking behaviours in the Shire. Given that there were not enough training data to build a robust prediction system, we built a pretraining model using data from other cities, such as the City of Melbourne, and performed a generative adversarial learning and domain adaptation to the new target city, in this case is the Mornington Peninsula. We showed that we could build an effective prediction model with the generative paradigm, using just six days of training data from Mornington Peninsula.
We also saw that human behaviour in cities, such as mobility behaviour and energy use patterns, are highly dynamic and prone to changes due to the externalities, such as extreme weather events, and the public health measures during COVID-19. Building AI-based predictive models that can still work effectively even during unprecedented events, edge cases, and unseen scenarios is critical.
By harnessing emerging technology including AI we can build long term, sustainable transport solutions and be truly forward thinking.
Data for resilience
Transport and infrastructure have in the past always involved a big build, but the question now is whether we need to create new assets or sweat existing assets more.
We need to look at whether it is more sustainable and efficient to make existing structures and networks more secure and resilient rather than embarking on new builds, harnessing insights from data. We must move away from the paradigm of the old.
Riding the wave
I love working with my students and working with new and exciting problems. That is why I am still in academia, with a strong interface with industry.
Throughout my career I have come to realise there are always a lot of projects of interest and people who will want a slice of your time. One of my previous mentors always said: ‘focus on the quality not the quantity’ and I think that applies to everybody. It’s not a number game. Look for the quality work and teams.
Secondly when you are building your career, you need to know when the wave is coming and be on top of the tide – it’s just like surfing. If you are waiting for the next wave to come, take the time to reflect and replan your next stages.