Webinars & Events

In engineering, a purely data-driven risk analysis is typically not an option because of rare failure events and the uniqueness of many systems. Therefore, engineering risk analysis typically requires a combination of models, data, and expert knowledge. At this BNMA July Webinar, Professor Daniel Straub will discuss that the BN is an ideal modeling tool in this context. It facilitates combining different models and consistently integrating information from various sources. Because of its graphical representation, it is also ideal for validation and communication. Daniel will show the potential and benefits of BN on several engineering risk analysis examples. He will also highlight some of the pitfalls and common misunderstandings, such as the confusion between diagnostic and causal reasoning, and discuss how to deal with them.

Daniel is Professor for engineering risk and reliability analysis at Technische Universität München (TU München). His interest is in developing physics-based stochastic models and methods for decision support and risk analysis of engineering systems. He uses BNs frequently for modeling and communicating uncertainties, both in research and practical applications.

Marco Scutari is a Senior Researcher at Istituto Dalle Molle di Studi sull’Intelligenza Artificiale (IDSIA), one of Switzerland's national AI research centres. Since completing his PhD in Statistics in 2011, he has held positions in Statistics Genetics (UCL), Statistics (Oxford), and Machine Learning (IDSIA) in the UK and Switzerland. His work focuses on the theory of Bayesian networks and their application to biological, environmental and clinical sciences, statistical computing, and software engineering for machine learning applications.

bnlearn (https://www.bnlearn.com/) is an R package that provides a complete solution for Bayesian network learning and inference. It started 15+ years ago as a small personal project to implement the algorithms Marco needed for his dissertation, which had no publicly available implementation. Today, bnlearn provides a comprehensive tool for both simulations and real-world data analysis. In this talk, Marco will discuss its design philosophy and his plans to tackle the main open problems in Bayesian network learning. He will use some of his ongoing data analyses as examples.

Archives are socially, politically and legally important to every functioning democracy. As more and more objects that need to be preserved are created and exist digitally (born digital) or are physical objects that have been digitised, understanding the risks to digital archives in order to put in place suitable mitigating strategies becomes more and more important. In this talk I will describe the Bayesian-network-based tool we developed in collaboration with The National Archive, UK, and its partner archives. The tool, DiAGRAM, (https://nationalarchives.shinyapps.io/DiAGRAM/) has been used by The National Archives to make a successful business case for substantial uplifts in funding in the UK Government spending review, was a finalist in the 2020 Digital Preservation Awards, and 2022 winner of The Decision Analysis Practice Award, sponsored jointly by the Decision Analysis Society and the Society of Decision Professionals, given annually to the best decision analysis application, as judged by a panel of members of both Societies.

Martine Barons is the Director of the Applied Statistics & Risk Unit at the University of Warwick, UK, a Chartered Mathematician, Vice President (Learned Society) of the Institute of Mathematics and its Applications, the UK representative of EU-MATHS-IN (which aims to leverage the impact of mathematics on innovations among stakeholders on a European level) a member of the advisory board to the Newton Gateway to Mathematics and the advisory board to the National Academy for Mathematical Sciences in the UK.

Koala conservation in NSW is complicated, and despite the popularity of the species, we are still don't have all the data we need to make a fully-informed choice on which actions we should prioritise. Luckily, there are some amazing experts in this area, with a wealth of knowledge and experience that they are willing to share. In this talk, UQ Graduate Emma Camus will explain how she used object-orientated continuous BNs to formalise expert knowledge on NSW koala management to create a practical decision support tool. Using this tool, we can run various management scenarios for the different regional landscapes within NSW, and show how a one-size-fits-all approach would have winners and losers across the different regions.

Emma is a Research Officer at Australian Council for Educational Research (ACER), where she designs, prepares, and conducts training in sampling for national and international research activities. Before working with ACER, Emma worked with Brisbane City Council to investigate the habitat of significant species in the Brisbane region. She graduated with honours from University of Queensland with a Bachelor of Environmental Management, majoring in Natural Systems and Wildlife.

As societies live and play closer to volcanoes there is a growing requirement for volcano monitoring agencies to provide quantitative information on the likelihood and impact of volcanic hazards that can be understood and utilized by landowners and managers, infrastructure managers, civil/federal aviation authorities, local people, and tourists. In addition, national health and safety legislation increasingly requires dynamic assessment of risk for employees, their subcontractors, and communication of the changing risk to other potentially affected parties. Eruption forecasting underpins such quantitative information and risk assessments.

Operational Scientist Yannik Behr will be telling us all about a Bayesian Network model to forecast the probability of eruption for Mt Ruapehu, Aotearoa New Zealand in collaboration with the New Zealand volcano monitoring group (VMG). Since 2014 the VMG has regularly estimated volcanic eruptions at Mt Ruapehu that impact beyond the crater rim. The Bayesian Network model structure was built with expert judgement based on the conceptual understanding of Mt Ruapehu and with a focus on making use of the long eruption catalogue and the long-term monitoring data. The model parameterisation was partly done by data learning, complemented by expert elicitation and is now implemented as a service to automatically calculate daily forecast updates.

The CoRiCal risk-benefit calculator (https://corical.immunisationcoalition.org.au/) has proven a popular tool for helping people making informed decisions about COVID-19 vaccinations. But did you know that it's hiding a Bayesian Network behind those charts? Our April ABNMS webinar will introduce you to the team behind CoRiCal and explain what it takes to use a Bayesian Network to turn journal articles and Government reports into a useful online decision support tool. We'll discuss the whole process, from model design and validation and finding the data, all the way up to risk communication and the logistics of getting it online.

While some projects are lucky enough to have a large dataset to work from, in many cases the reason a problem exists in the first place is because data is spread over different sources, in different formats, or exists only as expert knowledge. Bayesian networks (BNs) are your friend no matter which category you fall into.

In this webinar, Dr Helen Mayfield will go through some examples from Epidemiology in Fiji, policy design for the Great Barrier Reef, and making an informed choice on the COVID-19 Astra Zeneca vaccine to show how BNs can be used in decision support regardless of whether you have all the data, none of the data, or something in between.

This session will be delivered by Bruce G. Marcot, PhD, and will cover:

• The nuances and the tangled knots of what constitutes probability.
• When probability can, and cannot, be derived from frequency data.
• All Bayesian probability is conditional.
• Is probability absolute?
• Is probability real?
• A multiverse dive into the world of quantum Bayesian models -- question all you think you know!

If you want to be added to or opt out of our mailing list, please email us at contact@abnms.org.

Find out more about Bruce's research here!

This session is aimed at a general audience, and no previous experience in modelling or statistics is necessary. We will give you a high level introduction to Bayesian networks (BNs), answering questions such as:

• Why should I be interested to find out about BN modelling?
• What is a BN anyway?
• When is this a suitable modelling approach?

The "how" will not be covered in this session, but if you are interested in learning more, look out for the follow up webinar sessions.