Plenary Lectures
| Plenary Lecture 2 | Managing Uncertain Ground Truth Using Bayesian Machine Learning |
| Date / Time | 23 September 2019, Monday / 10:20 - 11:00 hrs |
| Speakers | Kok Kwang Phoon, Distinguished Professor and Vice Provost, Department of Civil and Environmental Engineering, National University of Singapore, Singapore
Alexander von Humboldt Research Award Winner 2018 Interim Director of Lloyd's Register Foundation, Institute on Public Understanding of Risk Academic Adviser to the Global Risks Report 2017 of the World Economic Forum Jianye Ching, Department of Civil Engineering, National Taiwan University, Taiwan |
The geotechnical engineering profession is highly successful in taking “calculated” risks by employing a hybrid strategy combining limited site and observational data, modelling, testing, precedents, experience, and judgment. However, there is no clear pathway to digitalization, which is transforming other industries in many fundamental ways. In a separate development, engineered systems are increasing in scale, complexity, interconnectivity, among others and the emerging resilience engineering paradigm in response to this challenge is to design for both expected and unexpected conditions. There is no precedent if a condition is truly unexpected.
The volume, variety, and velocity of data can only increase. A geotechnical engineer will soon be asking what to do with this deluge of data. In tandem, resilience engineering will require an adaptive rather than a precedence-based approach to risk management. This lecture advocates the need to view data as a primary asset that can directly improve our decisions and to develop “learning” algorithms that will allow decision making on quantities of interest to improve as more data accumulate. An example on estimation of soil/rock properties (a key design decision) is discussed. A Bayesian machine learning approach is adopted to characterize site effects and to estimate soil/soil properties under a set of general constraints abbreviated as MUSIC-X (Multivariate, Uncertain and Unique, Sparse, Incomplete, and potentially Corrupted data with variations in space/time, X). Data-driven decision making does not imply taking the engineer out of the entire life cycle management chain. It is intended to support rather than to replace human judgment.
