من اینجا رو دیدم اونو نوشتم:
We tried two different configurations of our architecture, one with dropout and no data-augmentation and another one without dropout and using data-augmentation. We name them Arch1 and Arch2 respectively
در کپشن شکل سه هم اینو نوشته
The non-dropout version is used in new tests with data-augmentation. The drop-out version was used on experiments without data-augmentation. The exclusion of dropout significantly increased the performance in augmented version, but may not perform the same for all scenarios, that’s the reason we included the base version with dropout. Another variation of the architecture with dropout is to use dropout layer after each layer, which we are currently testing.
من البته سریع خوندم ولی مقاله جالبیه.