Universal Domain Generalisation for Verisimilar Artificial Intelligence
Abstract
Australia's 2018 Robotics Roadmap suggests automation could add $2.2 trillion to the economy in 15 years, where developing secure Artificial Intelligence (AI) systems is one important enabler. A major hurdle is acquiring enough labelled data for training, hindered by costs and privacy issues, which affects AI adoption in sensitive sectors. This project enables efficient and effective AI with minimal data, by leveraging synthetic data and bridging simulation to reality gap. Focused on universal domain generalisation, this innovation could benefit businesses and start-ups by lowering costs and enhancing AI safety, and transform various sectors, from multi-media retrieval to autonomous driving, boosting Australia's economy and sustainability.
