Numerical Reproducibility based on Minimal-Precision Validation

Abstract : We proposed a new concept of weak numerical reproducibility and a systematic approach for it with a support of minimal-precision tuning and validation. We expect that the concept of weak numerical reproducibility covers most of the demands for reproducibility in computational sciences. Besides, the minimal-precision computing system can address the demands for accuracy, high-performance, and energy efficient computation as well; ultimately, if it has been realized with new hardware like FPGAs. Although the proposed system is still in development, some of key components are already available, and many of them have been developed by us.
Complete list of metadatas

https://hal.archives-ouvertes.fr/hal-02401878
Contributor : Fabienne Jezequel <>
Submitted on : Tuesday, December 10, 2019 - 11:21:37 AM
Last modification on : Friday, January 10, 2020 - 3:10:19 PM

Identifiers

  • HAL Id : hal-02401878, version 1

Citation

Toshiyuki Imamura, Daichi Mukunoki, Roman Iakymchuk, Fabienne Jézéquel, Stef Graillat. Numerical Reproducibility based on Minimal-Precision Validation. Computational Reproducibility at Exascale Workshop (CRE2019), in conjunction with the International Conference on High Performance Computing, Networking, Storage and Analysis (SC19), Nov 2019, Denver, United States. ⟨hal-02401878⟩

Share

Metrics

Record views

18