7/5/2022
•
EN
No, We Don't Have to Choose Batch Sizes As Powers Of 2
Challenges the common practice of using powers of 2 for neural network batch sizes, examining the theory and practical benchmarks.