Laurie Wired on Metaprogrammed Matrix Multiplication
This is interesting. These algorithms can be used to program silicon as well as general purpose machines. The hardest part seems to be parsing the tensor files! See https://github.com/LaurieWired/MatrixMultiplicationOptimization and https://github.com/mkauers/matrix-multiplication and the papers Flip Graphs With Symmetry and New Matrix Multiplication Schemes and Consequences of the Moosbauer-Poole Algorithms.
I wonder if people will start using techniques like this to devise algorithms customised to multiplication of particular kinds of matrices. See Scott Aaronson - Why Philosophers Should Care About Computational Complexity.
Subscribe to Laurie Wired.
Comments
Post a Comment