Optimizing GoogleNet using New Connections and Auxiliary Layers Information
Sepehr Maleki, Habib Izadkhah · 2022
GoogLeNet is a convolutional neural network developed by researchers at Google for image processing. In this study, we have optimized the GoogLeNet model. We divided inception blocks into two types of blocks and established new connections to improve the model accuracy. We created new links between the SoftMax layers to reduce information loss. These changes help us improve recognition accuracy. We compared the updated model performance with state-of-the-art architectures GoogLeNet, ResNet, and AlexNet on the CIFAR-10 dataset and achieved the best classification accuracy.