Skip to main content

Hindawi, one of the world’s largest open-access journal publishers, recently introduced its inaugural Article of the Year award. “Year after year, researchers from across the world publish exceptional work in Hindawi titles, helping to advance scientific knowledge for the betterment of society,” reads the announcement. “To celebrate Hindawi authors and their invaluable contribution to the advancement of science, we asked our Chief Editors to select the original research or review article published in 2020 which they consider to be exciting and impactful.”

Emiliano Tramontana, chief editor of the journal Scientific Programming, selected the article, “A Fortran-Keras Deep Learning Bridge for Scientific Computing,” by Jordan Ott, Mike Pritchard, Natalie Best, Erik Linstead, Milan Curcic and Pierre Baldi.

“The article represents a notable example of a technology helping developers tackle a relevant problem in everyday industrial applications, i.e. interfacing legacy systems with novel supports available,” says Tramontana. “The presented solution is valuable and reusable, and also offers suggestions on how to solve similar problems.”

The researchers introduced a software library, the Fortran-Keras Bridge (FKB), which connects environments where deep learning resources are plentiful with those where they are scarce. The source code is publicly available.

The work is the result of a multidisciplinary collaboration led by Jordan Ott, a computer science Ph.D. student in UCI’s Donald Bren School of Information and Computer Sciences (ICS). Ott and his adviser, Distinguished Professor of Computer Science Pierre Baldi, teamed up with Mike Pritchard, an assistant professor of Earth system science at UCI. The three of them collaborated with three researchers from Chapman University, including Erik Linstead, a former student of Baldi who is now associate dean and associate professor of Chapman’s Fowler School of Engineering.

As the authors explain in the paper, “the FKB library enables users to access many features of the Keras API directly in Fortran, including the ability to create custom layers and loss functions to suit their needs.” They present a case study involving multiscale physical simulations of the global atmosphere, demonstrating how the FKB library could produce considerable improvements in climate model stability.

— Shani Murray