Comparative Analysis of Recent Architecture of Convolutional Neural Network

Table 3

Architectural issues.

S.NO	Architecture	Benefits	Weakness
1	LeNet	Automatic learning of features reduces parameters and computation costs.	Large size filters; improper scaling of diverse classes of images.
2	AlexNet	Every type of feature extraction uses large as well as low-level filter sizes.	In the first and second layers, neurons are inactive.
3	ZfNet	Parameter tuning is introduced by visualizing the output of the intermediate layer.	Additional processing may be required for visualization.
4	VGG	The concept of the receptive field is introduced here. Proposes simple and homogenous topology.	Fully connected layers use high computational power.
5	Google Net	Multiscale filters are introduced. Dimensionality reduction is made by bottleneck layer.	Some information may be lost due to the application of the bottleneck layer.
6	Inception V3	Asymmetric filters and bottleneck are introduced for reducing the computation cost.	Designed architecture is complex.
7	Highway networks	A new training mechanism is introduced.	Complex design.
8	Inception V4	Multilevel feature extraction.	Architecture design which is complex may add up computation costs.
9	ResNet	Error rate is decreased, and residual learning concept is introduced.	Over-adaptation of hyperparameters for a specific task due to stacking of modules.
10	Deluge Net
11	Xception	Separable convolution is introduced. Application of cardinality for learning good abstraction.	High computational cost.
12	ResNeXt	Application of cardinality in each layer for diverse transformation, and grouped convolution is applied.	High computational power.
13	Dense Net	Ensures the maximum flow of data in layers, and redundant features are avoided in learning.	An increase in feature maps increases parameters.
14	Convolution block attention module	Applies global average and max pooling concept simultaneously and improves flow of information.	The computational load may increase.