GraphBLAS: Solving Graph Algorithms with Linear Algebra

Abstract

GraphBLAS is a C library written by Dr. Davis that allows users to easily represent graphs as sparse matrices. GraphBLAS also allows linear algebra operations on its graphs, so that users can develop graph algorithms in the language of linear algebra. Concluding that GraphBLAS is a more efficient and concise way of writing graph algorithms is important to academia, as it’d introduce a better approach for researchers and students to learn and write graph algorithms. The ability to write graph algorithms efficiently will allow researchers to test what they’re needing to do at a quicker pace. Instructors will also be able to teach and explain graph algorithms to their students in a way that they can easily grasp the material. In return, the students will get to learn the material in a new way and be able to test their understanding. My outcomes will further the validation and understanding of GraphBLAS as an alternative to regular graph algorithms. Furthermore, such graph algorithms will also allow for software developers in industry to write graph algorithms quickly. Such algorithms are crucial to various situations such as figuring out bots on Facebook and search results on Google. Kepner and Gilbert prefaced that graph algorithms “have become essential in controlling the power grid, telephone systems, and, of course, computer networks (xxv)”, further validating how impactful a new way to compute these algorithms could be. Previous research on this topic conducted by Buluc and Gilbert detail how to approach many different graph algorithms in the language of linear algebra. That research can be referred to gather information on how to better approach graph algorithms in GraphBLAS. We will be able to compare how easy and efficient it is to write such algorithms to the regular method using adjacency matrix or vertexes to test my research statement of GraphBLAS being a powerful and expressive way to develop graph algorithms.