報告題目：Bayesian Probability for Sensor Fusion and Pattern Recognition in Fusion Devices

報告時間：9月15日（星期一）上午9:30-11:30

報告地點：四號樓601會議室

報告人：Prof. Geert Verdoolaege

主持人：張洋 研究員

報告人簡介：Prof.?Geert?Verdoolaege?is?from?the?Ghent?University.?He?obtained?the?MSc?degree?in?Theoretical?Physics?in?1999?and?the?PhD?in?Engineering?Physics?in?2006,?both?at?Ghent?University.?His?research?activities?comprise?development?of?data?analysis?techniques?using?methods?from?probability?theory,?machine?learning?and?information?geometry,?and?their?application?to?nuclear?fusion?experiments.

報告簡介：

Fusion energy research can benefit greatly from modern data science methods, both for increasing the understanding of the underlying plasma physics and for optimizing the design and operation of fusion devices. From basic statistical techniques for model fitting, to Bayesian methods for probabilistic analysis of data from single or multiple diagnostics, to the latest machine learning techniques for anomaly detection and uncertainty quantification: the applications are numerous and the possible approaches originate from a broad range of subfields of the information sciences. In this talk, I will highlight a number of recent applications of Bayesian inference in fusion. I will start with sensor fusion, i.e. the systematic, joint treatment of data from multiple, heterogeneous diagnostics. In particular, Bayesian probability is seen to gracefully handle tomographic inversion using Gaussian process models. Opportunities for diagnostic design optimization and machine learning techniques for speeding up the inference process are also touched upon, with a view to real-time sensor fusion in future devices. I then proceed to applications of Bayesian inference and information geometry for pattern recognition in fusion data, concentrating on robust parameter estimation in complex, multi-machine data sets, as well as anomaly detection for predictive maintenance in fusion devices.