National AI month is about showcasing and celebrating Australia’s AI capabilities, talent and potential. Meet Flora Salim – an award-winning Professor in the School of Computer Science and Engineering, and the inaugural Cisco Chair of Digital Transport & AI, at UNSW Sydney – and learn about her research on machine learning for time-series and multimodal sensor data, and on trustworthy AI.

My research interests are in artificial intelligence and machine learning algorithms to model real world environment that produce multiple types of data: for example, time-series, text, audio, and video, especially from sensors.

The application areas of this research include individual activity and behaviour modelling, global scale climate and weather forecasting, urban-scale traffic and energy, and urban systems modelling.

Imagine sensors everywhere generating different types of data, capable of providing insights to solve different tasks, or forecasting ahead of emerging or unseen events.

My team and I have introduced self-supervised learning and pretraining methods for sensor data, which are important bases for data-efficient learning – a paradigm in deep learning that does not require a high volume of high-quality annotated data.

The pretrained model can be adapted and transferred to large-scale, new domain or tasks, demonstrating effective performance in unseen or unprecedented scenarios. These are a few examples of what I have been working on in this space.

In 2021, my postdoc Hao Xue and I pioneered the use of natural language processing to model time series data for forecasting task.

The latest iteration, PromptCast, uses natural language to prompt the time series forecasting as input and generate the corresponding output in natural language.

Professor Flora Salim. Credit: Supplied.

An example on how it could work is a prompt to ask, “What is the electricity consumption for the next week”. The answer is in complete sentences rather than just a number, e.g., “This client will consume 8337 kWh of electricity”.

The algorithm works on a wide variety of time series data, such as mobility/check-ins data, app usage behaviours, electricity consumption data, and weather data.

Recently, our machine learning and optimisation research on high dimensional sensor data from commercial buildings were linked with net zero initiatives, in the NSW Digital Industry Energy Flexibility (DIEF) project, an industry consortium led by CSIRO.

I am leading the Program on Analytics and Machine Learning for Carbon Performance in the new ARC Industrial Transformation Training Centre for Whole Life Design for Carbon Neutrality (2024-2029).

I am also leading the Mobilities Focus Area in the ARC Centre of Excellence for Automated Decision Making and Society (ADM+S).

In this multi-institutional national centre, we are focusing on developing trustworthy and responsible AI, acknowledging this is a multidisciplinary and sociotechnical effort.

I recently co-chaired the ADM+S Annual Symposium: Automated Mobilities at UNSW, which brought together innovative applications of automated decision-making across personal, shared, commercial, and public systems.

Read Professor Salim’s recent article in The Conversation about AI Superintelligence.

National AI month: Meet Professor Flora Salim