Marlin Receives $250,000 NSF Community Research Grant
UMass Center for Data Science professor Benjamin Marlin has received a three-year, $250,000 National Science Foundation grant from its Computer and Information Science and Engineering Community Research Infrastructure program to develop mResearch, a platform for reproducible and extensible research in mobile health.
The mResearch project will extend Marlin’s five-year collaboration with colleagues in the NIH-funded Center of Excellence for Mobile Sensor Data-To-Knowledge (MD2K) Center, a consortium of 11 universities and medical centers headquartered at the University of Memphis. MD2K’s current software platforms are used by scientists and engineers to collect health data from mobile sensors and devices like smart phones and to develop new algorithms to monitor health and wellness. Marlin says the new project will enhance MD2K’s existing software platforms with the goal of accelerating research in sensor design, mobile computing, privacy and data analytics.
Marlin is leading the development on campus of a platform for reproducible machine learning research with large-scale mobile sensor data. He says machine learning workflows in the mobile sensor data analysis space can be very complex, involving many layers of data transformations that use a collection of different tools and often include nuanced experimental procedures for evaluating new algorithms. Without a detailed description of the exact processes used, it can be very time-consuming to even approximately reproduce previous research results obtained by other teams, slowing overall progress in the field, he adds.
The mResearch platform will attempt to solve these issues by extending MD2K’s current data analytics platform with tools for compactly expressing complex sensor data processing and machine learning workflows. Once completed, this framework will allow researchers to publish their entire workflowfrom data cleaning and feature extraction to algorithm evaluation. By combining open data sets with published workflows, the goal is to enable exact reproduction of prior results with the click of a button, allowing the research community to spend more time on innovation and less time on re-establishing previously obtained baseline results, Marlin points out.