Deep learning, please tell me that it is a horse

In March 2016, the artificial intelligence robot Alpha Go defeated the five-part Li Shishi in the 4:1 record, and the whole world was shocked. Alpha Go introduces popular and effective deep learning algorithms and defeats humans with absolute advantage.

So what can deep learning technology do?

In the field of CV, whether it is industry or academia, on some large and open data sets including ImageNet, MSCOCO, etc., deep learning has the absolute advantage at the top of the leaderboard. Segmentation, restoration, classification, detection, and even image/video descriptions can achieve amazing results so far.

Other applications?

  • Use the neural network to automatically play Super Mario
  • Automatic translation
  • Automatically write ancient poems
  • Speech synthesis
  • Can do some simple optimization algorithms
  • Automatic programming
  • Song creation

There are also some interesting applications: zhihu.com/question/4756

Almost all applications, as long as the deep learning related ideas are applied reasonably, can always achieve surprising results. Some media are not to be outdone, and a large number of reports and comments on artificial intelligence to set off a new round of climax are constantly being introduced.

In fact, I came to pour cold water. Looking at the picture on the top, what is the thing in the green frame? We naturally know that it is a horse. And what does the neural network tell us?

This network has been trained for two weeks with some relatively large training data sets. It has reached the state-of-the-art level on some indicators, but still mistakes the horses running on the grasslands as birds. .

Whenever I see such a picture, I want to guard myself, and the world: although artificial intelligence has made great progress, it is far from the expected level. There is still a long time here. A long way to go.