Short answers, pulled from the story.
AlexNet is a convolutional neural network that sorts images into 1,000 categories, built for large-scale visual recognition. It became known for winning the 2012 ImageNet Large Scale Visual Recognition Challenge with a top-5 error rate of 15.3 percent, and it is regarded as an early landmark for deep convolutional networks at that scale.
AlexNet was developed in 2012 by Alex Krizhevsky, working with fellow graduate student Ilya Sutskever and Krizhevsky's PhD advisor, Geoffrey Hinton, at the University of Toronto.
A team called SuperVision submitted AlexNet to the ImageNet Large Scale Visual Recognition Challenge on the 30th of September 2012. It won with a top-5 error rate of 15.3 percent, more than 10.8 percent ahead of the runner-up.
AlexNet was trained on two Nvidia GTX 580 GPUs set up in Alex Krizhevsky's bedroom at his parents' house, over a period of five to six days.
AlexNet showed that a deep convolutional network could outperform hand-engineered feature methods like SIFT and SURF at large scale, a position that had been a minority view before its win. Its success came from combining a large labeled dataset, ImageNet, with GPU computing and improved training methods.
AlexNet contains eight layers, five convolutional and three fully connected, with 60 million parameters and 650,000 neurons. Because the network was too large to fit on a single GPU's memory, it was split across two Nvidia GTX 580 cards.