Comparing the Performance of Convolutional Neural Networks with and Without Batch Normalization
Kathayat Kalpana,
Akurathi Lakshmi Pathi Rao,
Thalla Umadevi
This research investigates the impact of Batch Normalization (BN) on the performance of Convolutional
Neural Networks (CNNs) by conducting a detailed comparative analysis of models with and without BN.
Using a standard CNN architecture, we evaluated the models across key metrics including accuracy,
loss, training time, and convergence rate, utilizing well-known datasets such as CIFAR-10 and MNIST.
The results demonstrate that the CNN with Batch Normalization consistently outperforms the non-BN
model, achieving higher accuracy, lower loss, and faster convergence. Additionally, the BN-enhanced
model requires significantly less training time, highlighting BN's role in improving training efficiency
and model generalization. This study underscores the critical benefits of integrating Batch Normalization
in CNNs, offering valuable insights for optimizing deep learning models in various applications.
