AlexNet: A Deep Convolutional Neural Network for Image Classification

AlexNet is a deep convolutional neural network (CNN) architecture that was developed by Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton in 2012. It was the winning entry in the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) in 2012, with a top-5 error rate of 15.3%, which was significantly lower than the second-place entry.

AlexNet consists of five convolutional layers, followed by three fully connected layers, and an output layer. The first two convolutional layers use a large filter size (11x11 and 5x5), with a stride of 4 to reduce the spatial dimensions of the input image. The remaining three convolutional layers use a smaller filter size (3x3), with a stride of 1.

The activation function used in AlexNet is the Rectified Linear Unit (ReLU), which is more computationally efficient than other activation functions such as the sigmoid function. The network also uses dropout regularization to prevent overfitting.

AlexNet was a significant step forward in the development of deep learning and CNNs, and it has since been widely used as a benchmark for image classification tasks.