A GPU Server for Integration of Machine Learning in Mathematics and Statistics Research and Training

用于将机器学习集成到数学和统计研究与培训中的 GPU 服务器

• 批准号: RTI-2021-00675
• 负责人: Zhang, Qingrun
• 金额: $ 10.92万
• 依托单位: University of Calgary
• 依托单位国家: 加拿大
• 项目类别: Research Tools and Instruments
• 财政年份: 2020
• 资助国家: 加拿大
• 起止时间: 2020-01-01 至 2021-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=717998
• 关键词: GPU; Server; Integration; Machine; Learning

We are requesting funds for a GPU server to provide critical computing power to support current and future research and training in Data Science for the Department of Mathematics and Statistics at the University of Calgary. The breakthrough of Machine Learning (ML) into both academic research, and industrial applications, has fundamentally changed the state-of-the-art of academic research, as well as ordinary citizens' lives. The integration of ML has become a trend in many fields. Moreover, there is also a growing population of trainees (among academic undergraduate and graduate students and professional continuing learners) who are enthusiastic to comprehend ML techniques in our research laboratories. Significant leverages for researchers to meet this need include modern ML libraries such as TensorFlow and PyTorch, which allow for the simple and prompt utilization of established ML tools. To embrace this exciting and ongoing data-oriented paradigm shift, the Department of Mathematics and Statistics has formed a Data Science group for ML research and training. It focuses on ML development tailoring to existing strengths in Biostatistics, Mathematical finance, High-dimensional statistical modeling, Neural network characterization, and Earth science. The applicants, an NSERC-funded diverse team of eleven (11) researchers representing this Data Science group, are integrating ML into their established research programs, supporting trainees from various backgrounds to give them experience and a knowledgebase aligning with their career goals. A roadblock to enacting the impact of this Data Science initiative is the lack of appropriate computational infrastructure tailoring to modern ML techniques. As such, we are requesting the support to acquire a GPU server with current mainstream specifications, which will allow for prompt deployment and execution of ML models. This equipment will be open to all researchers and trainees in the Math/Stats Department, providing them hands-on experiences in building and implementing ML models. Additionally, for non-ML computational tasks, the GPU server will also provide massive parallel computing power via a user-friendly interface, relieving researchers from writing code to manually coordinate multi-threads tasks (e.g., using OpenMP or MPI). The proposal is fully supported by the University of Calgary IT team, who will provide long-term storage, networking, maintenance, and user training. In summary, by supporting ML-focused and non-ML parallel computations, this infrastructure will benefit the research and HQP training in the Department, providing broader and better job opportunities for trainees in our graduate and undergraduate programs. When utilizing the equipment, actionable and measurable procedures will be implemented to facilitate a high degree of EDI in the department.

我们正在请求为GPU服务器提供资金，以提供关键的计算能力，以支持当前和未来的数据科学研究和培训，为卡尔加里大学数学与统计学系。机器学习在学术研究和产业应用上的突破，从根本上改变了学术研究的现状，也改变了普通公民的生活。机器学习的集成已经成为许多领域的趋势。此外，还有越来越多的受训人员（包括学术本科生和研究生以及专业继续学习者）热衷于在我们的研究实验室中理解机器学习技术。对于研究人员来说，满足这一需求的重要杠杆包括现代机器学习库，如TensorFlow和PyTorch，它们允许简单而迅速地使用已建立的机器学习工具。为了迎接这种令人兴奋和正在进行的以数据为导向的范式转变，数学与统计学系成立了一个数据科学小组，用于机器学习的研究和培训。它专注于ML开发，以适应生物统计学，数学金融，高维统计建模，神经网络表征和地球科学的现有优势。申请人是nserc资助的一个由11名研究人员组成的多元化团队，代表了这个数据科学小组，他们正在将机器学习整合到他们已经建立的研究项目中，支持来自不同背景的学员，为他们提供经验和符合他们职业目标的知识库。实现这一数据科学倡议影响的一个障碍是缺乏适合现代ML技术的适当计算基础设施。因此，我们请求支持获得符合当前主流规范的GPU服务器，这将允许快速部署和执行ML模型。该设备将向数学/统计系的所有研究人员和学员开放，为他们提供构建和实施ML模型的实践经验。此外，对于非ml计算任务，GPU服务器还将通过用户友好的界面提供大量并行计算能力，使研究人员不必编写代码来手动协调多线程任务（例如，使用OpenMP或MPI）。该提案得到了卡尔加里大学IT团队的全力支持，他们将提供长期存储、网络、维护和用户培训。总之，通过支持以机器学习为中心和非机器学习并行计算，该基础设施将有利于部门的研究和HQP培训，为我们的研究生和本科生课程的学员提供更广泛和更好的工作机会。在使用这些设备时，我们会实施可操作和可衡量的程序，以促进部门内高度的电子数据交换。