Home

Numerical Linear Algebra meets Machine Learning

Goal

This project tries to use a neural network to indicate which solver will be the fastest on a dense matrix. It has been built for the practice of software engineering project at the KIT.

Technology

For the neural network we used Keras. To indicate the speed of the solver we used the Ginkgo library, which allows GPU-accelerated solvers. The matrixes for testing/training are downloaded from the SuitSparse Matrix Collection with the SSGET command line tool by the Ginkgo group.