Jakub Lengiewicz

Abstract: Data from populations of systems are increasingly prevalent. Infrastructure continues to be instrumented with sensing systems, emitting streams of telemetry data with complex interdependencies. Data-centric monitoring procedures tend to consider these assets/datasets as distinct – operating in isolation and associated with independent data. In contrast, this work looks to capture the statistical correlations and interdependencies between models that represent groups of systems. Utilizing a Bayesian multilevel approach, the value of data can be extended, since population data can be considered as a whole, rather than constituent parts. In particular, domain expertise and knowledge of the underlying physics have the potential to be encoded within the multilevel structure: at a system, subgroup, or population level – as well as between systems.

Additional material:

• papers that most of the talk focused on: https://arxiv.org/abs/2305.08657 and https://arxiv.org/abs/2204.12404
• learning the latent noise model: https://arxiv.org/abs/1404.5443
• implementing a similar model in Stan: https://avehtari.github.io/casestudies/Motorcycle/motorcycle.html

Lawrence A. Bull is a research associate in the Engineering Dept. at the University of Cambridge, within the Computational Statistics and Machine Learning group. He researches statistical methods for monitoring telemetry data from systems and infrastructure, working closely with the Cambridge Centre for Smart Infrastructure and Construction (CSIC).