Research from Dr Ali Zia is helping develop AI systems that better understand the complex structure of real-world data.
“Most AI models learn by examining individual pieces of information, such as pixels in an image, or by comparing two pieces at a time,” he explains. “But many real-world and scientific problems are more complex than that.”
Dr Zia's research develops higher-order and structure-aware AI methods that allow models to consider how multiple elements interact and how information is organised across space, time, and different sensing modalities.
“Higher-order representation capture relationships among several elements at once, while structure-aware methods preserve important properties such as connectivity, topology, spatial organisation and consistency across different views of the same problem.”
Dr Zia says this approach could improve AI in situations where accurate decisions are critical.
“In medical imaging, structure-aware AI could help detect abnormalities while preserving important anatomical boundaries, such as the shape and connectivity of an organ or lesion,” he says.
“It could also help autonomous vehicles and robots combine information from radar, cameras and other sensors more consistently, improving their ability to recognise pedestrians, vehicles and road structures, even in poor weather or low visibility.”
Dr Zia’s next step will be to extend structure-aware and higher-order representations to scientific and agricultural sensing.
“In agriculture, this approach could combine images, spectral measurements, environmental sensors and observations collected over time to detect early crop stress, disease, or pest activity before the symptoms become clearly visible.”
“The longer-term goal is to translate these advances into practical decision-support systems for agriculture, healthcare, autonomous sensing, and industrial monitoring, while also developing collaborative projects with scientific and industry partners.”
