Computing Workstations for Deep Learning on Graph-Structured Data

在NSERC RTI的支持下，我们将建立图神经网络实验室（GNN-Lab），这是温莎大学的第一个深度学习实验室，特别关注将深度神经网络应用于图结构数据。如果没有图形处理单元（GPU）加速神经网络工作负载，将耗时从数月减少到数天甚至数小时，大大提高研究的可行性以及以有效方式进行和传播尖端研究的能力，GNN研究是棘手的。温莎大学已经承诺通过最近（2020年）聘请一位领先的GNN早期职业研究员来支持GNN研究的发展，这使得建立GNN实验室变得至关重要和紧迫。该提案要求8个支持GPU的计算机工作站引导GNN-Lab，支持UWindsor当前和未来的GNN研究。根据加拿大政府于2019年推出的1.25亿美元泛加拿大人工智能战略，与加拿大的研究机构密切合作开发和领导人工智能，我们的GNN实验室配备了所需的工作站，点燃了温莎在推动人工智能发展方面的贡献，同时帮助当地的生态系统和工业合作伙伴。具体来说，GNN-Lab的总体目标是在图论，时间序列分析和生物信息学之间的接口进行前沿研究，以克服当前研究的局限性，i）信息检索，ii）社会网络分析，iii）通过三个主要的协同研究计划A，B和C的药物设计。GNN-Lab将推动GNN上较少探索的人工智能的进展，在研究计划A和B中使用社会信息（Social IR）构建智能搜索引擎进行信息检索，并在研究计划C中使用药物再利用进行生产性制药管道。使用所要求的支持GPU的工作站，HQP将获得GNN在社交网络分析，信息检索和药物再利用方面的应用培训并变得熟练，使学生处于图形结构数据处理知识的最前沿。GNN-Lab将与行业合作伙伴保持一致的合作关系，将研究成果转化为工作原型，为HQP提供独特的机会，让他们看到自己的努力在行业中得到应用，同时为未来的职业发展提供见解。然后，HQP可以立即被搜索引擎就业市场吸收，包括谷歌以及制药行业，包括辉瑞加拿大公司。

With NSERC RTI's support, we will establish the Graph Neural Network Lab (GNN-Lab), the first deep learning laboratory at the University of Windsor with a special interest in applying deep neural networks on graph-structured data. GNN research is intractable without graphics processing units (GPUs) that accelerate neural network workloads, reducing elapsed time from months to days or even hours, dramatically increasing the feasibility of the research and ability to conduct and disseminate cutting-edge research in an efficient manner. The University of Windsor has already committed to supporting the growth of GNN research through the recent (2020) hire of a leading GNN Early Career Researcher, making the establishment of a GNN-Lab critical and urgent at this time. This proposal requests 8 GPU-enabled computer workstations to bootstrap GNN-Lab, supporting current and future GNN research at UWindsor. In line with 2019's launched of $125M Pan-Canadian AI Strategy by the Government of Canada to develop and lead AI in close collaboration with Canada's research institutes, our GNN-Lab fueled with the requested workstations, ignite Windsor's contribution in pushing forward the state-of-the-art in AI, and meanwhile helping local ecosystems and industrial partners. Specifically, the overarching objective of GNN-Lab is to perform cutting-edge research at the interface between graph theory, time series analysis, and bioinformatics to overcome the limitations of current studies in i) information retrieval, ii) social network analysis, and iii) drug design through three prime synergistic research programs A, B, and C. GNN-Lab will push forward the progress of less explored AI on GNN that finds immediate applications in building intelligent search engines for information retrieval using social information (Social IR) in research programs A and B, and productive pharmaceutical pipeline using drug repurposing in research program C. Using the requested GPU-enabled workstations, HQP will acquire training in and become adept at the application of GNN on social network analysis, information retrieval, and drug repurposing, putting students at the cutting edge of graph-structured data processing knowledge. GNN-Lab will see consistent engagement with industrial partners where research findings will be translated into working prototypes, providing a unique opportunity for HQP to see their efforts be applied in the industry while providing insights into the next career steps. HQP can then immediately be absorbed by the search engine job market, including Google as well as the pharmaceutical industry, including Pfizer Canada Inc.