Statistical Inference in High-Dimensional Discrete Spaces: A Bayesian Approach

Luis Carvalho

Fellow
Hariri Institute for Computing
Boston University

Abstract: Many problems from varied fields can be seen as statistical inference tasks in large dimensional discrete spaces, such as classification, clustering, or variable selection. These problems are often challenging due to two main factors: few observations relative to the number of variables, and structured spaces that invalidate usual independence assumptions. In this talk we present three applications that illustrate these issues and discuss models and an estimator that aim at addressing these drawbacks. These applications feature large datasets and come from social sciences, genetics, and environmental sciences. The estimator aims to summarize well the complex space of solutions while achieving computational feasibility in the face of “big data”. This is joint work with Hunter Glanz, Ian Johnston, and Lijun Peng.

Bio: Prof. Luis Carvalho joined the Department of Mathematics and Statistics at Boston University in 2009. He has two MSc in Transportation Engineering and Computer Science, and a PhD in Applied Math from Brown University.

His research interests cover Bayesian and computational statistics with a focus on high-dimensional discrete inference. His research finds applications in social sciences, genetics, and environmental sciences; in particular, he works with community detection in large social networks, genome-wide association studies, and land cover classification in global scale remote-sensed satellite images.