Seminar 2 in ourdata science seminar series between the Institute of Statistical Mathematics in Japan and the University of Bristol.
There are 2 talks in seminar 2 of the series ‘Compositionally-warped additive mixed modeling: application to COVID19 data in Japan’ by Daisuke Marakami & ‘Active pooling design in group testing based on Bayesian posterior prediction’ by Ayaka Sakata.
Compositionally-warped additive mixed modeling: application to COVID19 data in Japan
An increasing number of non-Gaussian geospatial data is becoming available. At the same time, the size of spatial data rapidly grows together with the development of sensing technology. Given such background, this study develops a flexible additive mixed modeling approach for large non-Gaussian data. The development is done by combining an additive mixed model (AMM), which accommodates spatial and other effects, with the compositionally-warped Gaussian process (CWGP) estimating the shape of data distribution that can be either Gaussian or non-Gaussian possibly have skewness, fat tail, and other properties. Monte Carlo experiments shows that the proposed approach improves coefficients estimation accuracy without losing computational efficiency relative to the linear AMM. The developed approach is then applied to a spatiotemporal analysis of COVID-19 in Japan, investigating the influence of age, people flow, and other factors on the infection.
Speaker: Daisuke Murakami, Assistant Professor, Department of Statistical Data Science, The Institute of Statistical Mathematics.
Daisuke Murakami received a PhD in Engineering from the University of Tsukuba, Tsukuba, Japan, in 2014. From 2014 to 2017, he was a Research Associate at the National Institute for Environmental Studies, Japan. In 2017, Daisuke Murakami became an Assistant Professor at the Institute of Statistical Mathematics, Japan. His research interests include spatial and spatiotemporal statistics, quantitative geography, and urban analysis.
Active pooling design in group testing based on Bayesian posterior prediction
For identifying infected patients in a population, group testing is an effective method to reduce the number of tests and correct test errors. In group testing, tests are performed on pools of specimens collected from patients, where the number of pools is lower than that of patients.
In this talk, Bayesian inference and a corresponding belief propagation (BP) algorithm are introduced to identify the infected patients from the results of tests performed on the pool, estimating the prevalence in the prior information. Further, an adaptive design method of pools based on the predictive distribution is proposed in the framework of Bayesian inference. The proposed method results in more accurate identification of the infected patients, compared with the group testing performed on random pools determined in advance.
Speaker: Ayaka Sakata, Assistant Professor, Department of Statistical Data Science, The Institute of Statistical Mathematics.
Ayaka Sakata received her PhD degree from the University of Tokyo in 2011. She spent four years as a postdoctoral researcher at Tokyo Institute of Technology and RIKEN, and is currently an associate professor in the Institute of Statistical Mathematics.
She was awarded the Young Scientist Award from the Ministry of Education, Culture, Sports, Science, and Technology, Japan, in 2019. Her research interests include statistical physics of random systems and statistics.
More about the seminar series: Institute of Statistical Mathematics, Japan and University of Bristol, UK Modern large-scale datasets often exhibit rich geometric structure that is often overlooked in classical methods of analysis, and yet may provide the answer to a number of scientific questions. This series included theoretical and methodological developments in the area of geometric data analysis, and also practical applications in diverse fields, such as earth sciences, biology, and material science.
This seminar series was partially supported by Daiwa Anglo-Japanese Foundation