MRI: Acquisition of a Heterogeneous GPU Cluster to Facilitate Deep Learning Research at UMBC

点击翻译按钮获取中文摘要

This project acquires an instrument to pursue large-scale, data-intensive, end-to-end research, aiming to service several fields that include natural language processing (NLP), robotics, computer vision (CV), computer graphics, cybersecurity, medical analysis, and other heavily statistical areas. Utilizing very large data sets and performing intensive computation using Graphics Processing Units (GPU), the work focuses on vastly expanding the GPU computation power, storage, and access to data. This effort aims to reflect the discipline-wide shift within engineering towards model and system-building that require GPU computation and a continued focus within some computer and information science and engineering (CISE) disciplines towards deep learning and big data in need of big storage arrays, high memory servers, and horizontal scaling capabilities. The instrument contributes in preparing students with the skill to use clusters and other tools for handling large problems with the help of the cluster.Researchers will be able to tackle problems in various areas with the heterogeneous architecture of the cluster, since cluster-based computing can increase availability, reliability, and scalability. Moreover, performance on tasks that can be parallelized might be improved. Deep learning techniques will likely improve performance in predictive modeling. In turn, the cluster facilitates research across multiple discipline, it enables work in robotics, healthcare, medicine, as well as trust and fairness in machine learning. The proposal supports young faculty, women, and underrepresented minority groups such as UMBC (University of Maryland-Baltimore County), Meyerhoffer, and CWIT (Center of Women in Technology).This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.