https://www.acm.org/articles/people-of-acm/2019/tim-davis
> I picked Gaussian elimination. "That will be quick," I thought. Not! I got started on matrices in 1985 and I'm not done with Gaussian elimination yet.
> GraphBLAS is a community effort, including industry, academics, and government labs, that is working to design a library that can implement graph algorithms based on sparse linear algebra over semirings. There are lots of great graph libraries out there that don't exploit the linear algebraic abstraction, but most of what they do can be viewed as matrix operations on adjacency matrices. GraphBLAS makes the connection to linear algebra explicit.
> GraphBLAS gives the graph algorithm developer the power and abstraction of linear algebra, with all its advantages (associative and distributive laws, AB=(B'A')', and so on). One level of breadth-first search, for example, is a single sparse matrix-times-sparse vector multiply, with no loss of asymptotic efficiency. Google's entire PageRank computation can be done with a handful of iterations around a single call to GraphBLAS, using what I call the "PageRank semiring."