DAIRNet AI Symposium 2024

Dr Mel McDowall is the Director of the Defence AI Research Network (DAIRNet), an initiative of the Department of Defence and hosted by the University of South Australia. This draws on her experiences in transdisciplinary research and operation roles. Having worked in many industries, such as clinical, agriculture and Defence, Mel has experience with many steps in the “bench to bedside/farm-gate/capability” journey. This includes basic and applied research, business improvement and governance, advocacy, contract management, territory management, and clinical support.

Tony is Director of Lockheed Martin Australia’s (LMA) Advanced Systems & Technologies (AST), Lockheed Martin’s first international multidisciplinary R&D Laboratory. Prior to his role at LMA, Tony was with the Defence Science and Technology Group for twenty-eight years. His last position was Chief of the National Security and Intelligence, Surveillance and Reconnaissance Division. In that position he was privileged to lead a world-class team that created and transitioned multiple advanced ISR capabilities to a range of defence and intelligence operations for Australia and her allies. LMA’s AST was established with the focus on operational transition and the organisational ethos of impactful R&D that were foundational to those successes.

Ralph Gailis received his BSc (Hon) in physics and mathematics at the University of Melbourne in 1992. He completed his PhD in theoretical physics at the University of Melbourne in 1996, as well as a postdoc in 1997-1999, in the area of cosmology.
Ralph began working in Defence Science and Technology in 1999, contributing to a broad range of research topics including Chemical, Biological and Radiological (CBR) hazard modelling / threat analysis, diseases surveillance and epidemic modelling, and further work in data fusion and machine learning.

Dale A. Lambert has assumed a number of roles within Defence Science and Technology in the Australian Department of Defence, including Chief of Information Sciences Division; Chief of Cyber and Electronic Warfare Division; Chief of National Security and Intelligence, Surveillance and Reconnaissance Division; Director General of Science Strategy and Policy; and Research Leader of Intelligence Analytics.

As Head of the Department of Aerospace Engineering at RMIT University (Australia’s leading Aerospace department), Professor Jennifer Palmer leads a team of 25 staff and 85 research students specialising in aerospace engineering and aviation. Her current research is focussed on aerial autonomous systems, particularly teaming and swarming operations, along with the design and trialling of novel air vehicles. Jennifer has more than 20 years’ experience in industry, government, and academia and was previously Technology Program Director at the Trusted Autonomous Systems CRC.

Associate Professor Belinda is Deputy Director of the UniSA Industrial AI Research Centre, the Australian Chair of the Technical Advisory Panel for the Defence AI Research Network and is an adjunct faculty member at Johns Hopkins University. She has over 25 years’ experience working with Defence, Government and Industry, including US Homeland Security, on military and intelligence applications, utilising data analytics for modelling, decision-making and visualisation. She has led multi-stakeholder projects for Defence and Government, as well as working as a consultant for state Government.

Presentation title: Neural memory architectures for scalable reasoning over temporal multimodal data.

Abstract: Multimodal data – like text, audio, video, and sensor feeds – offer powerful insights but are challenging to analyse effectively. This challenge is particularly acute when working with data from multiple sensors and network systems that generate vast amounts of information over time. Despite these challenges, such data analysis can deliver significant benefits. For example, it can enhance cyber threat detection, enable real-time situational awareness, and support intelligent decision-making by analysing network patterns and system behaviours. To address these challenges, we present UNITED – a scalable AI framework that integrates self-supervised learning with deep reasoning capabilities. UNITED is designed to process complex, time-varying data from multiple sources using advanced encoding techniques and memory systems. This enables accurate analysis and prediction across extended timeframes. UNITED implements three specialised architectures: 1) A system which organises information in a variable-size hierarchical memory; 2) A system featuring a controlled memory module for complex reasoning tasks, and 3) A system which uses multiple specialised memory modules for handling diverse data types. This general framework advances our ability to process complex multimodal data for real-world applications across Defence, industry, and research domains.

Presentation title: Sequential Decision-making for Robots Operating in Non-Deterministic and Partially Observable Worlds.

Abstract: In recent years, robotics hardware has advanced tremendously, with increasingly affordable humanoids, quadrupeds, telepresence robots, and many more. Despite these advances, developing autonomous or semi-autonomous robots that can reliably, efficiently, and safely operate in our environments remains an open problem. Key to this difficulty is the ubiquity of uncertainty. These robots must compute strategies to achieve their goals even when the outcomes of their actions are uncertain, their sensors and perception systems are erroneous, and the environments they operate in are dynamic and only partially observable. Moreover, they must ensure safety for both the robots and the humans around them. However, the technology that enables robots to efficiently construct effective strategies in the presence of a wide variety of uncertainty is still lacking. In this talk, I will present some of our work in developing such a technology, specifically in our work on the Partially Observable Markov Decision Processes (POMDPs) —the general and principled framework for sequential decision-making under uncertainty. I will also present how this technology can be applied for safety assurance of autonomous systems.

Presentation title: From Noisy Data to Early Detection: Preliminary Insights and Challenges in Analysing Wearable Biosensor Signals from a Time-Locked Immune Challenge Study.

Abstract: To explore the potential of consumer wearables for early infection detection, we conducted a unique, time-locked immune response study involving over 100 participants. Each participant was monitored using wearable biosensors that tracked signals such as heart rate variability, temperature, and respiratory rate, alongside daily questionnaire responses. Baseline data were collected over a ten-day period, followed by a vaccination, with monitoring continued for an additional three days to identify potential physiological markers of immune activation. This talk will discuss the distinct challenges involved in analysing biosensor signals from wearables, including data variability, input errors, and the complexities of data cleaning and processing. We will present preliminary findings from our ongoing analysis. Additionally, we will outline our plans to refine these analyses, address data quality issues, and develop methodologies to more accurately identify reliable early indicators of immune response from wearable data. By sharing both our findings and challenges, we aim to enhance the understanding of wearable-based health monitoring and examine its potential for real-time infection detection.

Presentation title: The Standard of Article 36 reviews: Connecting military and civilian AI governance.

Abstract: Before any means or method of warfare is deployed in armed conflict, the legality of its use should be confirmed by legal review, at the stage of “study, development, acquisition or adoption.” This review should ensure that new means and methods of warfare comply with international law. International law is comprised of treaties, international custom, and general principles of law. This legal review ensures that all new or modified means or methods of warfare comply with general legal principles, for example, the requirement that they are not “of a nature to cause superfluous injury.” Examples of treaty-based obligations include prohibitions on poisons, or biological or chemical weapons, just to name a few. “Nano” is a prefix meaning “a billionth” (a factor of 10−9 of a meter) derived from the Greek νᾶνος, meaning, “dwarf”. By way of illustration, a nano-sized object is to an apple, what an apple is to the size of the earth. Or to give another example, one nanometer particle could fit approximately 80 000 time across a human hair.

Presentation title: GenAI for Defence: From Cybersecurity to Enhanced Decision-Making.

Abstract: This talk will explore the application of GenAI within potential defence contexts ranging from cybersecurity to decision-making. It will cover how LLMs can be utilised to bolster cybersecurity across a variety of threat scenarios, including software supply chains. Moving beyond cybersecurity, the talk will introduce reference designs for improving decision-making capabilities through the integration of third-party LLMs. Key focuses include scaffolding architectures, guardrails, design options, and the development of use case-specific learning structures that operate outside the model. By harnessing these elements, the talk aims to demonstrate how adaptable and secure, GenAI-driven systems can be tailored to meet the unique challenges of defence operations.

John Thangarajah is a Professor of Artificial Intelligence and the Director of Research for the Centre for Industrial AI Research & Innovation, at RMIT University. He was previously the head of the Computer Science and Software Engineering discipline leading and managing the education programs and research. 
 
John is an internationally recognized research leader in the technical area of Autonomous Multiagent Systems. He has a sustained record of accomplishment in research publications and research funding from both government and industry sources. In particular, he has established strong collaborations with the Australian Defence sector around the modelling and simulation of complex behaviours.