paint-brush
An Overview of the Data-Loader Landscape: Numerical Results Cont.by@serialization
193 reads

An Overview of the Data-Loader Landscape: Numerical Results Cont.

Too Long; Didn't Read

In this paper, researchers highlight dataloaders as key to improving ML training, comparing libraries for functionality, usability, and performance.
featured image - An Overview of the Data-Loader Landscape: Numerical Results Cont.
The Serialization Publication HackerNoon profile picture

Authors:

(1) Iason Ofeidis, Department of Electrical Engineering, and Yale Institute for Network Science, Yale University, New Haven {Equal contribution};

(2) Diego Kiedanski, Department of Electrical Engineering, and Yale Institute for Network Science, Yale University, New Haven {Equal contribution};

(3) Leandros TassiulasLevon Ghukasyan, Activeloop, Mountain View, CA, USA, Department of Electrical Engineering, and Yale Institute for Network Science, Yale University, New Haven.

A. NUMERICAL RESULTS CONT.

In this appendix, we include a collection of plots for which we did not have space in the core pages of the article.


Figure 11. Comparing the impact of batch size in CIFAR10 with a single GPU.


Figure 12. Comparing the impact of batch size in Random with a single GPU.


Figure 13. Comparing the impact of batch size in CoCo with a single GPU.


Figure 14. Comparing the impact of the number of workers in CIFAR10 with a single GPU.


Figure 15. Comparing the impact of the number of workers in Random with a single GPU.


Figure 16. Comparing the impact of the number of workers in CoCo with a single GPU.


Figure 17. Comparing the impact of batch size in CIFAR10 with multiple GPUs.


Figure 18. Comparing the impact of batch size in Random with multiple GPUs


Figure 19. Comparing the impact of batch size in CoCo with multiple GPUs.


Figure 20. Comparing the impact of the number of workers in CIFAR10 with multiple GPUs


Figure 21. Comparing the impact of the number of workers in Random with multiple GPUs.


Figure 22. Comparing the impact of the number of workers in CoCo with multiple GPUs.


Figure 23. Comparing the impact of batch size in CIFAR10 with a single GPU while filtering.


Figure 24. Comparing the impact of batch size in Random with a single GPU while filtering


Figure 25. Comparing the impact of batch size in CoCo with a single GPU while filtering.


Figure 26. Comparing the impact of the number of workers in CIFAR10 with a single GPU while filtering


Figure 27. Comparing the impact of the number of workers in Random with a single GPU while filtering.


Figure 28. Comparing the impact of the number of workers in CoCo with a single GPU while filtering.


This paper is available on arxiv under CC 4.0 license.