The AlexNet architecture won the 2012 ILSVRC by a large margin, with a top-5 test error rate (that is, the correct label must be in the top 5 predictions) of 15.4%. By comparison, the second-best submission, which wasn’t based on a deep network, trailed at 26.2%. This was a defining moment in the history of computer vision: the moment when the community started to realize the potential of deep learning for vision tasks. That leap was followed by constant improvement, with more modern architectures and training methods getting top-5 error rates as low as 3%.
This paragraph is mentioning a moment that makes the community realize the potential of deep learning. Are there any other similar defining moments in the history of deep learning?