Author here: neural-fortran is a parallel microframework for building and training fully-connected neural nets of arbitrary shape and size. This is the companion code to the Chapter 6 of my book with Manning (https://www.manning.com/books/modern-fortran) in which I explain Fortran derived types and collective subroutines.
neural-fortran may be interesting to Fortran programmers who want to learn how to implement a neural net (I think I did!), or for machine learning practitioners interested in learning more about Fortran.
This is currently a proof-of-concept, but may have potential for use in production. I don't know how fast it is, but it's about 3 times as fast as Michael Nielsen's numpy implementation which I used as reference during development.
neural-fortran may be interesting to Fortran programmers who want to learn how to implement a neural net (I think I did!), or for machine learning practitioners interested in learning more about Fortran.
This is currently a proof-of-concept, but may have potential for use in production. I don't know how fast it is, but it's about 3 times as fast as Michael Nielsen's numpy implementation which I used as reference during development.
Comments and questions are very welcome!