Table 3 Evolution of key CNN... | Emerald Publishing

Table 3

Evolution of key CNN architectures and their features

Year	Architecture	Developers	Key features
1998	LeNet-5	Yann Lecun	7-layer CNN for handwriting and character recognition, includes convolutional, average pooling, fully connected layers and a softmax classifier
2012	AlexNet	Alex Krizhevsky; Ilya Sutskever; Geoffrey Hinton. Research	Deeper and wider than LeNet, uses data augmentation and dropout for overfitting reduction, consists of convolutional layers, max-pooling, fully connected layers and softmax
2014	VGGNet	Visual Geometry Group; University of Oxford	Uniform architecture with 3 x 3 convolutional layers, stride 1, SAME padding and 2 x 2 max-pooling. Contains 16 convolutional layers
2014	GoogLeNet	Szegedy et al., 2015	Introduced Inception Module, reduced parameters, replaced fully connected layers with average pooling, total 22 layers
2015	ResNet	He et al.	Introduced “skip connections” and “identity shortcut Connection,” facilitating direct gradient backpropagation and Supporting increased depth without performance degradation

Year	Architecture	Developers	Key features
1998	LeNet-5	Yann Lecun	7-layer CNN for handwriting and character recognition, includes convolutional, average pooling, fully connected layers and a softmax classifier
2012	AlexNet	Alex Krizhevsky; Ilya Sutskever; Geoffrey Hinton. Research	Deeper and wider than LeNet, uses data augmentation and dropout for overfitting reduction, consists of convolutional layers, max-pooling, fully connected layers and softmax
2014	VGGNet	Visual Geometry Group; University of Oxford	Uniform architecture with 3 x 3 convolutional layers, stride 1, SAME padding and 2 x 2 max-pooling. Contains 16 convolutional layers
2014	GoogLeNet	Szegedy et al., 2015	Introduced Inception Module, reduced parameters, replaced fully connected layers with average pooling, total 22 layers
2015	ResNet	He et al.	Introduced “skip connections” and “identity shortcut Connection,” facilitating direct gradient backpropagation and Supporting increased depth without performance degradation