Unit 7.3 Convolutional Neural Network Architectures

References

• Krizhevsky, Sutskever, and Hinton (2012). ImageNet classification with deep convolutional neural networks

• Canziani, Paszke, Culurciello (2016). An analysis of deep neural network models for practical applications

• Simonyan and Zisserman (2014). Very deep convolutional networks for large-scale image recognition

• Szegedy, Liu, Jia,Sermanet, Reed, Anguelov, Erhan, Vanhoucke, Rabinovich (2014). Going Deeper with Convolutions

• He, Zhang, Ren, and Sun (2016). Deep residual learning for image recognition

What we covered in this video lecture

Coding convolutional neural networks like AlexNet and VGG16 is relatively simple, thanks to PyTorch’s convenient abstractions. But, on the other hand, developing convolutional neural network architectures that work well is also an art. So, in practice, it’s recommended to build on existing, successful architectures that researchers spent significant time developing. In this lecture, we reviewed the major convolutional network milestones and architectures, including AlexNet, Inception, VGG, and ResNets.

Additional resources if you want to learn more

Among many others, popular convolutional architectures include MobileNet (particularly small) and EfficientNet v2 (small but also state of the art).

Watch Video 1 Mark complete and go to Unit 7.4 →