Two Queenslanders Successful in 2022 AI4DM Round 3
AI4DM Round 3 – QDSA is pleased to announce that there have been two (2) successful Queensland recipients in the latest round of the Artificial Intelligence for Decision Making Initiative (AI4DM) in 2022.
Successful recipients receive fixed funding of $30,000 per project. They are eligible to be invited to join a national network focused on developing AI (Artificial Intelligence) and ML (Machine Learning) technology through the Defence AI Research Network (DAIRNet).
Project prototypes developed as a result of the funding may also be considered for progression through the Defence Artificial Intelligence Centre (DAIC) or Defence and university sectors.
Funding such as the AI4DM grants is pivotal in the success of developing technologies and instrumental in supporting the growth of sovereign capability in Australia.
The Queensland Defence Science Alliance is the facilitator for the Queensland AI4DM program, administered on a state-by-state basis and combines funding from multiple Commonwealth agencies.
Past recipients of AI4DM have represented many vital areas of technological advancement, including deep learning, investment portfolio optimisation, multimodal data extraction, semantic integration of uncertain and evolving data and more.
Let’s look at the scope of projects by the 2022 successful recipients.
Queensland’s 2 Recipients for 2022
1. Professor Ravinesh Deo – University of Southern Queensland
Recipient: Professor Ravinesh Deo, Program Director – Master of Science (MSCN) including GDSC and GCSI, University of Southern Queensland (Project Lead) and Mr Chris Davey, Doctoral Researcher
Project title: Identification of significant field types and boundaries in unfamiliar network protocols: Sequential field tokenisation and type classification
Snapshot of project description: The project is part of a larger research effort to identify and characterise the potentially adversarial events in command-and-control systems at a network protocol level. It is grounded within the cybersecurity domain and aims to extract useful information about the structure and type of data from often proprietary network protocols. Such information is used as part of a larger data processing pipeline that characterises activities and events occurring within the networked control system.
The project seeks to achieve this by developing a data-driven methodology for the training of a byte-level sequence-to-sequence neural network model. Toward this end, the project team aims to evaluate the efficacy of such a model on “known” protocols and its capability to generalise to “unknown” protocols.
Potential outcome: Ideally, at the outcome of this project we aim to provide a transferrable methodology and set of processes that will enable our collaborators to further their own research objectives.
The developed methodology, documentation and demonstration capability will contribute to the larger project of our collaborators within Defence whose aim is to automatically monitor and identify potentially adversarial events both historically and online.
2. Professor Clinton Fookes, Queensland University of Technology
Recipient: Clinton Fookes, Professor in Vision & Signal Processing and co-Director of the SAIVT group (Signal Processing, Artificial Intelligence and Vision Technologies) within the School of Electrical Engineering & Robotics of the Faculty of Engineering, QUT (Project Lead), Prof Sridha Sridharan (Co-investigator) QUT, Dr Tharindu Fernando (Research Fellow), QUT, and Dr Dana Michalski (Collaborator), Intelligence Analysis Branch, Defence Science and Technology Group.
Project title: A Multi-Modal Deep Generative Framework for Video-Based Identification
Snapshot of project description: This project aims to extract multiple sources of information from video to boost the performance of person identification. While there have been great gains in facial recognition technology to identify people in video, the success of such a tool relies completely on the visibility of the face within the acquired image or video footage which is not always available.
This project looks at other complementary sources of information within the video which can help improve person identification. Such sources include logos, clothing appearance, tattoos, a person’s trajectory, voice, and other soft biometric features such as their age and gender.
The project is using generative adversarial learning and multi-task learning approaches to incorporate these additional sources of information. The project also addresses the challenge of domain mismatch, where a particular modality can visually look quite different when captured under different viewpoints or different environmental conditions.
Potential outcome: There have been great gains in biometrics such as facial recognition technology over recent years, but in some situations, the face of the person of interest is simply not available. As such, there are major technical challenges hindering the policing, forensics, defence and law enforcement responses that may be required. The technology developed in this project can assist in the more effective identification of children and people of interest through the use of these other complementary sources of information contained within video.
More Defence Capability Funding Opportunities
QDSA is committed to connecting defence capability funding opportunities to eligible recipients. Programs such as AI4DM provide prospects with a foot in the door to actualise concepts and ideas that have been many years in the making and could be the foundation of ground-breaking advances globally.
QDSA Upcoming Events and Opportunities
QDSA has many upcoming events and opportunities, with more being added each week. Want to find out what is on the horizon? Check out our News section on our website here or subscribe to our e-newsletter here.
The Queensland Defence Science Alliance (QDSA) is a university-led initiative to grow and connect an agile Defence innovation ecosystem, leveraging Queensland’s strengths, to deliver trusted solutions to meet Defence requirements.