Researchers will build secure AI technology to handle rapidly changing environments with funding from the Australian Defence Force (ADF).
UNSW researchers have been awarded $3,220,633 by the ADF's Advanced Strategic Capabilities Accelerator's Emerging and Disruptive Technologies program . to develop a new generation of multimodal foundation models (MFMs) - powerful AI systems that can take in information from multiple sources, combine and interpret that information and self-update as situations change.
AI expert and project lead Professor Flora Salim, from UNSW's School of Computer Science and Engineering and Deputy Director (Engagement) at the UNSW AI Institute , said the project would focus on creating systems that can perform in highly dynamic defence environments.
"We are working on building MFMs that can adapt to rapidly changing conditions, protect against attacks and manipulation, and handle emerging inputs such as audio, video and sensors. By combining these inputs through advanced data fusion and post-tuning techniques of Large Language Models (LLMs) and Vision Language Models (VLMs), we will develop AI systems that continually learn and self-update, to provide a reliable understanding of complex situations," she said.
The models are designed to operate in some of the most challenging environments faced by defence teams, where data is unpredictable, access is intermittent or limited and decisions must be made swiftly.
We will develop AI systems that continually learn and self-update, to provide a reliable understanding of complex situations.
A key innovation will be developing new security defences for multimodal AI that are not currently available. This includes input and output filtering, along with adversarial training to help the system learn to identify and manage attacks and hostile environments. By combining robust performance, cross-modal reasoning and built-in security, the project promises to set a new standard for trustworthy, resilient AI.
"This project underpins UNSW's commitment to translating advanced research into outcomes that genuinely help our partners, whether they're in academia, industry, government or the broader community," Head of the School of Computer Science and Engineering Professor Arcot Sowmya said.
The Director of UNSW AI Institute, Dr Sue Keay, said the project demonstrated Australian researchers' strong capabilities in developing home-grown sovereign AI.
"I'm proud to work with the chief investigators on this project, Prof. Salim, Prof. Salil Kanhere, Dr Rahat Masood and Dr Aditya Joshi, and their very talented teams. I look forward to watching their work integrated into critical ADF infrastructure," she said.
