Deep Learning Platform for Large-Scale Representation Learning and Visual Understanding
Abstract
Requested Equipment. This project requests equipment, a high-performance graphics processing unit (GPU) cluster with eight BIZON G7000 servers, to support the proposed research and educational activities. Each server owns 1024 GB memory and 8 Nvidia RTX 2080 Ti GPUs. The requested equipment offers an extremely useful computing resource that will have a lasting impact on research and education. Research Objectives. The requested high-performance GPU cluster will support a research project on knowledge-guided scene graph generation and reasoning, which has been proposed to ARO recently. The key idea is to achieve knowledge-guided visual scene understanding, by generating and reasoning over scene graphs with the help of external knowledge. We will also design novel scene graph learning approaches to achieve visual understanding in open and dynamic environments. As considerable computational costs are expected in the proposed research, the requested GPU cluster will facilitate the proposed research activities and accelerate the research progress. Moreover, the requested GPU cluster will support the PI and collaborators at the University of Georgia (UGA) to explore sever emerging research topics on visual representation learning, including transferable feature learning for visual domain adaptaiton, multimodal fusion for action recognition and prediction, self-supervised visual representation learning, and semantic-aware visual learning applications. The PI is willing to leverage the proposed equipment to produce more preliminary results for future DoD white papers and full proposals. In summary, the requested GPU cluster will support the proposed research project, help establish new research capabilities on large-scale visual learning, and contribute to the effective implementation and timely success of the research milestones. Educational Benefits. Extensive educational activities at UGA will be benefited from the requested GPU cluster. Over 600 students at UGA are taking courses on computer vision, machine learning and artificial intelligence every year. Many of them are willing to engage to the DoD-relevant research topics on deep learning, video analysis, intelligent systems, information fusion, etc. The requested GPU cluster will provide critical computing resources to these students through course projects, faculty mentored research for undergraduate students, and graduate research projects. Along with the proposed research activities, the requested GPU cluster will also bring DoD-relevant research opportunities to students who are the next generation of scientists and engineers. In addition, the project is also integrated with extensive educational and outreach activities to share code and data, organize workshops and tutorials, create graduate courses, and motivate K-12 students in research.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 25, 2021
- Source ID
- W911NF2110028
Entities
People
- Sheng Li
Organizations
- Army Contracting Command
- The University of Georgia
- United States Army