engleski

A Brief Note on the Influence of Storage Choices on Machine Learning Algorithm Training Times

Training times of ML algorithms is one of their biggest pitfalls, with a large number of researchers and developers working on ways to speed up the process. This paper attempts to determine the influence of used storage, within a realistic environment – foregoing bloated datasets and models, on the training times. The research utilizes two models, on two different architectures, across four different storage options: Solid State Drive (SSD), Hard Disk Drive (HDD), RAMDISK and network accessed storage. The results show that there is not a large difference in training times, when observing realistic cases and suggests that a significant influence on training times is not exhibited by storage.

Artificial Intelligence, Execution Timing, Machine Learning, Performance Profiling

nije evidentirano