CS231N-CNN Architectures

Common used architectures

AlexNet:

VGG16:

VGG19:

GoogleNet:

Inception:

Performance of CNN models:

The deeper model performs worse, but it’s not caused by overfitting.

Hypothsis: the problem is an optimization problem, deeper models are harder to optimize.

Solution: ResNet.

ResNet:

Residual block:

Complexity

The best in the left: Inception-v4: ResNet+Inception
For the right:

• VGG:Highest memory, most operations
• GoogleNet: Most efficient
• AlexNet: Smaller compute, still momory heavy, lower accuracy
• ResNet: Moderate efficiency depending on model, highest accuracy

Forward pass time and power consumption

Reference

http://cs231n.stanford.edu/slides/2018/cs231n_2018_lecture09.pdf