AlexNet

AlexNet is a convolutional image classification network architecture that won the ImageNet visual recognition challenge in 2012. It was designed by Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton. The network achieved a top-5 error of 15.3%, more than 10.8 percentage points below the runner-up. It is considered one of the most influential papers in computer vision, leading to many more published papers that use CNNs to accelerate deep learning.

The network has eight layers: 5 convolutional layers (some followed by max-pooling layers) and also 3 fully connected layers. It can be depicted by:

$$(CNN\to RN\to MP{)}^2\to (CN{N}^3\to MP)\to (FC\to DO{)}^2\to Linear\to softmax$$

where:

• CNN: convolutional layer (ReLU)
• RN: local response normalization
• MP: maxpooling
• FC: fully connected layer (ReLU)
• Linear: fully connected layer (no activation)
• DO: dropout layer

Sources

• https://en.wikipedia.org/wiki/AlexNet
• https://www.analyticsvidhya.com/blog/2021/03/introduction-to-the-architecture-of-alexnet
• https://proceedings.neurips.cc/paper_files/paper/2012/file/c399862d3b9d6b76c8436e924a68c45b-Paper.pdf