A PhD student from The University of Queensland’s School of Electrical Engineering and Computer Science has been awarded the prestigious 2024 Google PhD Fellowship in Security, Privacy, and Abuse Prevention.
Zihan Wang was among only five early-stage postgraduate researchers from across Australia to receive the 2024 award, and the first Australian student to be recognised in the Security and Privacy track since the program was established in 2017.
The fellowship award will help support Zihan’s research, along with a Google mentor to provide guidance on technological expertise and career development.
Under the supervision of UQ’s Associate Professor Guangdong Bai and Senior Research Scientist Dr. Jason Xue from CSIRO's Data61, Zihan’s work aims to enhance the protection of deep AI learning and support researchers and technology companies to safeguard their intellectual property and promote responsible use of AI.
"Usage control is a fundamental problem in Artificial Intelligence," Zihan explains.
“It involves ensuring that once an AI model is shared or deployed, it cannot be misused or exploited without permission.”
His research aims to tackle the problem of companies losing control over AI models after sharing them with partners or integrating them into devices, leading to risks like unauthorised copying, intellectual property theft, or malicious applications.
"Existing protection methods can be costly and not entirely effective," Zihan said.
“Techniques like watermarking only help identify misuse after it has occurred, they don't prevent it."
To address this challenge, Zihan has developed a method that enables model owners to limit a model's functionality before sharing it.
"My work aims to intentionally reduce a model's capabilities so that only authorised users with a special access key can unlock its full potential," he says.
“For example, a company can distribute a basic version of a sophisticated AI model, keeping advanced features locked.
"By controlling access to the model's full capabilities, we protect the creator's rights and minimise the risk of unauthorised use.”
His work aims to not only secure valuable technology but also promote safer collaboration between companies.
"Ultimately, we want to make sharing AI technologies more secure and efficient," Zihan says.
"By ensuring models are used responsibly, we can encourage innovation while protecting everyone's interests."
Looking to the future
Having commenced his postgraduate studies in 2023, Zihan aims to complete his PhD by 2026.
"The joint support from The University of Queensland and CSIRO's Data61 provides an excellent environment for young researchers like me," he says.
"I'm very grateful for this award and for the opportunity to work with Professor Bai and Dr. Xue."
Zihan is also a visiting scientist at CSIRO's Data61 and has participated in several industry placements. He aspires to continue working in AI security as a researcher in the private sector or academia.
