Parallel and Distributed Computing

点击翻译按钮获取中文摘要

The performance of many large scientific and data analytics codes for applications in machine learning and physics simulations depend on the existence of high-performance frameworks. Often, efforts in developing mathematical algorithms are disjoint from performance engineering efforts that build the scientific software. The emergence of new applications and stupendously large data has rendered the classical approaches to building high-performance software inadequate in many situations. This research program demonstrates that large-scale problems can be solved significantly faster when mathematical methods and high-performance software are developed jointly because data dependencies in software programs often originate from the mathematics that formulates the problem.