Microsoft "Xiaoice" Dog and Artificial Neural Network (III)

around Microsoft "Xiaoice" the Dog has a "fog" (ie difficult to understand), until this year 2 Month 6 Microsoft technology Executive Sun Jian published an article confirming that the "Xiaoice Dog" feature is Microsoft's proprietary "Artificial Neural network" ( ANN to the vast number of users to provide a network service, so far, this piece of "fog" is gradually eliminated.

But what is "artificial neural network"? As the name implies, artificial neural network is composed of "artificial neurons" of the network, or artificial "neural network"? We use the previous understanding, that is, the artificial neuron (Artificial Neuron) network. So, what is "artificial neurons"? What is the network of "Xiaoice dog"? How many of these artificial "neurons" are there?

It is known that the neurons in the human brain are approximately spherical, with many elongated "axons" around them ( Axons ), the shape is complex. Artificial "neuron" shape is not spherical, but a weight, the uniform size of the "box", convenient installation in the machine room in the large cabinet. We might as well assume that "Xiaoice dog" uses 10 cabinets, each with a single, artificial "neuron" inside each cabinet. This hypothetical table is easy for us to imagine and explain.

What is the man-made "neuron" in these cabinets, we don't say for the moment. We first give the cabinet number: the first cabinet to accept the external signal "input", the tenth cabinet responsible for the system's "output", the other is the "middle-tier" cabinet. "Xiaoice Dog: The process is equivalent to the dog's photo into the first cabinet, after automatic processing, in the tenth cabinet to output the recognition results, that is, the dog category name."

the question now is, how do these "neurons" act together, pass on information, process messages, and eventually realize the task of "recognizing dogs"? The reality is that these artificial "neurons" are not connected using tens of thousands of thin fibers, but instead use "IP addresses" to connect to each other. There is nothing to be done about the following.

the specific identification process (or algorithm is), the dog's initial photo input into the first cabinet, upload to the second cabinet, the machine identified some characteristics of the dog, very vague, continue to upload to the third cabinet, the other part of the dog features identified, the image is gradually clear up, so continue, like "winding" (convolution) action, has been "winding" to the tenth cabinet, the dog's face revealed the "truth", recognition task completed. Ah, it turns out to be the most popular image and speech recognition technology in the world today: convolutional Neural Network (CNN)! "Convolution" (convolution) is the constant "winding" meaning. How to "winding"? and listen to tell.

Shimen 7 Month - Day

Copyright NOTICE: This article for Bo Master original article, without Bo Master permission not reproduced.

Microsoft "Xiaoice" Dog and Artificial Neural Network (III)