ATHENA: A Fortran package for neural networks

Abstract

In the landscape of modern Fortran programming, there exists a compelling need for neural network libraries tailored to the language. Given the extensive set of legacy codes built with Fortran, there is an ever-growing necessity to provide new libraries implementing on modern data science tools and methodologies. Fortran’s inherent compatibility with high-performance computing resources, particularly CPUs, positions it as a language of choice for many machine learning problems. The vast amount of computing capabilities available within current supercomputers worldwide would be an invaluable asset to the growing demand for machine learning and artificial intelligence. The ATHENA library is developed as a resource to bridge this gap; It provides a robust suite of tools designed for building, training, and testing fully-connected and convolutional feed-forward neural networks.