Impact of Batch Normalization and Spatial Dropout on VGG-11, NiN, and GoogLeNet

We study the effects of batch normalization and spatial dropout on the performance of VGG-11, Network in Network, and GoogLeNet architectures. We incorporate these regularization techniques into each model and evaluate their impact on classification accuracy and average loss. Experimental results show that spatial dropout has a limited influence on performance metrics, whereas batch normalization consistently improves model performance. Moreover, combining both methods does not yield significant gains beyond those achieved by batch normalization alone.