DURIP: Heterogeneous GPU Cluster to Support Deep Learning Research in Computational Vision, Audition and Neuroscience
Abstract
Statement of Work:The PIs are requesting a GPU cluster, high-capacity memory, servers, high band-width communication, and peripheral equipment. The cluster will be located at the University of Maryland and shared by collaborators at Stanford University and Virginia Tech. Researchers in all three universities have current DoD grants including MURIs, multi-investigator, and single investigator grants. This cluster, for research in deep learning, is expected to lead to significantly enhanced new algorithms that are able to tackle challenging problems that arise in large visual and acoustic data sets collected from moving platforms to inspire a number of new research directions in the area of opportunistic sensing, which is ofhigh interest to defense and security.Objective:Purchase a GPU cluster and peripherals for research in deep learning applied to many areas including computervision, computational acoustics for multi-modal surveillance and human-robot collaboration.Approach:The research supported by the requested equipment includes: (a) automated understanding of scenes, activities and events from multimodal surveillance and monitoring, primarily images/video and sound; (b) robot perception; (c) computational models of human behavior and decision making for use by autonomous agents; and (d) human-robot collaboration. All of these problem areas are of high importance in defense and security. The proposed equipment housed in the University of Maryland (and shared with Stanford and Virginia Tech for collaborative research) will significantly enhance the research and educational experience of graduate students who are currently supported by DOD and IARPA projects in the labs of the three participating institutions. The research areas supported by theproposed equipment are of high importance to DoD; these include computer vision, computational acoustics, and neuroscience necessary for automated understanding of the environment with multi-modal surveillance and humanrobot collaboration.Overall Merit and ONR Mission/Relevance:The proposed equipment will support research in disciplines related to the ONR focus areas of Autonomy, and Information Dominance. The proposed equipment is expected to support significant advances in computer vision and computational acoustics. The PI and co-Pis are leading researchers engaged in collaborative research in automated scene and event understanding, and human-agent collaboration, which is of high importance to DoD. All three universities have strongprograms in science and engineering and the requested equipment will further enhance their research and educational programs.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 30, 2016
- Source ID
- N000141612909
Entities
People
- Larry Davis
Organizations
- Office of Naval Research
- United States Navy
- University of Maryland