Download PDF

Science-Integrated Statistical Learning: interpretable statistical modeling for effective decision-making

Organiser

Dr Simon Mak

Participants

Dr Simon Mak (Chair)

Prof. Chih-Li Sung (Presenter/Speaker)

Causal regression for interpretable prediction for fusion design

Oksana Chkrebtii (Presenter/Speaker)

Likelihood-free inference for dynamical systems via reconstruction maps

Mike Grosskopf (Presenter/Speaker)

mgGP: Mesh-grouped Gaussian process for partial differential equations systems with uncertainty quantification

Daniel Williamson (Discussant)

Conference

64th ISI World Statistics Congress

Category: International Society for Business and Industrial Statistics (ISBIS)

Abstract

Advances in sensing technologies, experimental techniques and scientific computing have led to an increasing accessibility of high-quality data. As such, statistical learning models have become an integral tool for investigating key scientific and industrial questions. There are, however, fundamental limitations. Existing models are often black-box and lack scientific interpretability. High-quality scientific data can also be expensive, resulting in small sample sizes for model training, and thus poor predictive models with high uncertainty. This session focuses on novel developments in science-integrated statistical learning theory and methods, which tackle these issues by integrating scientific domain knowledge (e.g., mechanistic models, differential equations, guiding principles) as prior information for statistical model building. The session will feature speakers who have worked in this area for a broad range of applications in science, engineering and industry, followed by a discussant.