Browse Teach Curate

Multi-Armed Bandit Autotuning of HPC Application Parameters on Edge Devices

Research contribution: LASP, a lightweight autotuner that casts HPC application parameter selection as a multi-armed bandit and runs low-fidelity trials on resource-constrained edge devices to discover configurations that are then transferred to full HPC clusters, reducing tuning time and energy while respecting edge compute limits.

KITT Text

HPC Application Parameter Autotuning on Edge Devices: A Bandit Learning Approach Abrar Hossain1

This performance-engineering research paper (cs.PF) introduces LASP (Lightweight Autotuning of Scientific Application Parameters), a method for tuning the configuration parameters of high-performance…