SC21 Proceedings

The International Conference for High Performance Computing, Networking, Storage, and Analysis

Reproducibility in HPC: Passing Fad or a Work in Progress?

Moderator: Daniel Reed (University of Utah)

Panelists: Beth Plale (Indiana University), Fernando Perez (University of California, Berkeley), Lorena Barba (George Washington University), Tanu Malik (DePaul University), Tom Scogland (Lawrence Livermore National Laboratory)

Abstract: This is an era of disruptive technologies and changing science needs. Artificial intelligence (AI), science at scale through Jupyter, performance portability and growing energy consumption awareness suggest dynamism in how rigor in science must continue to evolve. But is enhancing rigor through efforts to reproduce publications at submission time a fad or is it a work in progress? SC demonstrates reproducibility at the point of manuscript submission, and 50% of paper submission authors (SC21) seek to have their results evaluated for ease of reproduction. An SC20 survey, however, found that a large majority of respondents feel that transparency is a more important outcome than is reproducibility. This panel brings together SC reproducibility experts with highly regarded disruptive technologists who represent changing science needs. The panel promises to hash out the issues for enhanced rigor in high-performance computing science going forward.


Back to the Panel Archive Listing