In fact, Microsoft on July 14, 2014 first disclosed an ambitious research program "Adam Plan" ("PROJRCT Adam"), the project is a research project on computer vision, which is later "Xiaoice dog" function. "Adam Plan" uses the ordinary PC processor (that is, CPU) as the "Avatar" of the artificial neurons, the Schondo between these processors established a "logical connection", you can efficiently train the system (that is, the system continuously input the reference image). For example, identify the species of the Welsh short Legged Dog (corgis) (breed). This kind of dog is very fat, body pudgy, "charmingly naïve", is a kind of dog, have different breed. The general person is not clear what the specific species, but the "Xiaoice" can be distinguished. In this respect, people are inferior to machines. In fact, the training Picture (resource) of the "Xiaoice Dog" feature comes from the Imagenet website. The site has 14 million images, divided into 22,000 sub-categories, including hundreds of dog "photos" (database). It is not easy for the machine to control the picture "dog". How to train the machine is the key. We should note that the "Xiaoice dog" function, will be further improved in August this year, able to speak with the voice, and will "associate", say thoughtful words. This shows that the Microsoft development team has used other training materials "training" system to make the system more intelligent, "understand the matter." As it goes on, the consequences are terrible. "Xiaoice" robots are able to "know the cat" and ... until they can "meet people". July 29, Win 10 released, means that some cold "AI big cabinet" (machine brain) directly (or secretly) connected to the Internet, and we do not feel, eventually realize the human and machine "dances." Alack! Description: Artificial neurons are complex and ready to be elaborated in another article. Shimen July 20
Copyright NOTICE: This article for Bo Master original article, without Bo Master permission not reproduced.
Microsoft "Xiaoice" Dog and Artificial Neural Network (V)