Dr Stewart Worrall
Research fellow at Australian Centre for Field Robotics
The efficient and safe mobility of people and goods in urban areas is a challenging problem that becomes more difficult with increasing population density. Congestion, accidents and pollution are highly undesirable for the financial, environmental and general well-being of the population. The automation of vehicles has long been considered part of a “smart cities” solution to this problem, working towards a utopian, sci-fi vision of the future. These technologies have been long promised, but seem to be always just out of reach. This presentation will highlight some of the key challenges in vehicle automation. The application of machine learning as a “black box” approach does not easily fit with a highly safety critical engineered system. The talk focused particularly on aspects of autonomous systems that are appropriate for deep learning, but also where these techniques are not appropriate given our current understanding of the problem & what ML techniques might be applied to address these issues.
4th year Ph.d student Australian Centre for Field Robotics
Over the last few years, deep learning techniques for image based semantic segmentation have been demonstrated to produce remarkable results for applications in intelligent transportation systems. However, the issue of the robustness has recently been recognised as a major challenge for the massive deployment of this new technology. In particular, for autonomous vehicles, any erroneous classification could potentially lead to catastrophic consequences. In this talk, Wei explained the main challenges we are facing at the moment when applying semantic segmentation to autonomous vehicles. Wei presented her proposed method to validate the robustness of semantic segmentation by automatically generating ground-truth labels. This method can be used in most real-world driving scenarios without the time and expense of using humans to generate labels by hand.
Chair of Standards Artificial Intelligence Committee
Aurelie is a member of ‘The Australian AI Working Group’, established by Standards Australia with the aim to become the preeminent forum for exploration and discussion of AI by engaging both government and industry; using the global IEEE s’ working group for Standard P7000 ‘Model Process for Addressing Ethical Concerns During System Design’; and The European AI Alliance, a forum established by the European Commission. Separately from her initiatives around Ethics and AI, she is a practising lawyer with over 10 years’ experience in Financial Services.
Verge Labs, Director, Data science leader
Anthony Tockar is director and cofounder at Verge Labs, an AI company focused on the applied side of machine learning. A jack-of-all-trades, he has worked on problems across insurance, technology, telecommunications, loyalty, sports betting and even neuroscience. He qualified as an actuary, then moved into data science, completing an MS in Analytics at the prestigious Northwestern University.
Professor Maurice Pagnucco
Deputy Dean (Education), Engineering and the Head of the School of Computer Science and Engineering at UNSW
Professor Maurice Pagnucco, the Head of the School of Computer Science and Engineering at UNSW talked about computational complexity theory - one of the most popular topics in computer science today.
Computational complexity theory focuses on classifying computational problems according to their inherent difficulty and relating these classes of problems to each other.
Professor Claude Sammut
Professor in the School of Computer Science and Engineering, University of New South Wales
Machine learning has had many recent successes, but current popular methods have their limitations, including requiring many training examples and a lack of transparency. Classical AI techniques can help to learn from a small number of examples and make the result more explainable. In particular, classical AI has well-developed methods for representing and reasoning about background domain knowledge. In this talk, Professor Claude Sammut described the hybrid learning methods they have used in robot learning that combine symbolic learning and planning with numerical methods. Examples of such robot learning systems include learning a bipedal walk, a robot learning to use tools and a rescue robot learning to traverse rough terrain.