Decision superiority is fundamental to success in military endeavours. Artificial intelligence is increasingly relevant to achieving this decision superiority. Differentiable decision trees (DDTs), also known as propositional logic networks (ProLoNets), allow human subject matter expertise to be represented as decision trees that can be encoded as artificial neural networks, and which are amenable to deep reinforcement learning (RL) algorithms. Consequently, DDTs allow us to warm-start RL and have a high degree of explainability. In this presentation, the fundamental ideas behind DDTs will be introduced and demonstrated in the context of a Kaggle Competition based on Halite: a video game consisting of 4 competing players where the objective is to mine the most halite. Players do so by building and controlling spaceships. Whilst this toy problem is a far-cry from Defence applications, the lessons learned about building and deploying DDTs are insightful.
About the speaker:
Dr Daniel Salmond leads the Information Warfare Command and Control capability within the Cyber and Electronic Warfare Division of the Defence Science and Technology Group. He completed a PhD in Telecommunications Engineering in 2013 with the University of South Australia, and a Bachelor of Science (Physics, Physical Mathematics) with Griffith University in 2000. On any given day, he may consider himself a mathematician, physicist or engineer. He has worked in Australia and the United States on researching and developing machine intelligence algorithms for military systems. He is currently focussed on the research and development of cognitive systems for defending the integrity of the Australian information environment.
23 / 06 / 2021 - 05:30 pm to 06:30 pm
- Member Rate: $0.00 ($0.00 excl. GST)
- Technical Society Member Rate: $30.00 ($27.27 excl. GST)
- Student: $0.00 ($0.00 excl. GST)
- Non-Member Rate: $30.00 ($27.27 excl. GST)
Engineers Australia Member Services
1300 653 113
Maximum CPD Hours