Integration and Benchmarking of State-of-Art Convergence Accelerators of the EM Algorithm

• Activity Number: 458
• Type: Topic Contributed
• Date/Time: Wednesday, August 1, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing
• Author(s): Jennifer Bobb*+
• Companies: Harvard School of Public Health
• Keywords: EM algorithm ; Convergence acceleration ; Benchmark studies

Abstract

In recent years several algorithms to accelerate slowly-convergent fixed-point iterations such as the EM algorithm have been introduced. However, widespread use and direct comparison of the algorithms across a broad range of applications has been challenged by the lack of a common implementation framework. In this talk we present "turboEM", an R package that integrates four state-of-the art accelerators in a common environment. A major utility of this integration is that it allows the user to compare the performance of each acceleration technique for the problem at hand. Through examples, we demonstrate tools for conducting benchmark studies and visualizing results, thereby facilitating comparison of multiple algorithms.