Dedicated high-powered workstation for machine and deep learning in genomics and industry

点击翻译按钮获取中文摘要

***The Department of Mathematics and Statistics at Memorial University of Newfoundland is in the process of significantly strengthening its research endeavours in Data Science. The recent formation of the 12 member Data Science Research Group (https://www.mun.ca/math/dsci), as well as the proposal of a one year MSc in Data Science are just two of our most recent collective efforts to put data science in Atlantic Canada on the map. The present application for infrastructure support is one undertaking in this regard. The requested infrastructure will primarily support the NSERC-funded programs of research of the co-applicants, but will also support the work of the other 10 members of the Data Science Research Group. The infrastructure will enable significant collaboration, co-supervision of HQP, and will allow us to provide state-of-the-art training in highly marketable skills to HQP in this increasingly vital field.*********Our current needs to deploy deep learning or preprocess data to use in machine learning and bioinformatics require a powerful dedicated workstation to allow parallelization, testing, tune-up, and verification. Ordinary machines are simply not capable of tolerating such processing and memory requirements. We have tried to use Compute Canada resources, however due to queuing and wait listing, this resource is not adequate particularly at the prototyping stage where we have to change and tune parameters. We have access to the IBM LSF 10.1.0.6 machine through Memorial University's department of Genetics (www.med.mun.ca/CHIA) and to the cluster Torngat, which is shared among the members of the Centre for Numerical Analysis and Scientific Computing. Unfortunately, both machines are not suitable for the proposed research projects due to RAM constraints and limitations in terms of their GPU capabilities.******The problems we pursue involve enormous, and highly confidential datasets that cannot be efficiently analysed with the infrastructure currently available at MUN. In order to use real datasets to build and test accurate models (for disease detection, weather prediction, and anomaly detection); remain competitive in the data science field; publish in top-tier journals or conferences; attract the best HQP; and to provide HQP with state-of-the-art training, while protecting the integrity of the data, the computational power of the requested equipment is vital.