Deep Learning Popular Science

First, it's up to the father of Ai, Turing.

Turing once had a dream uninstall "computer and Intelligence" (1950) article, if one day, the computer can do, across the wall, you do not know the opposite and you communicate is a person or computer, then this computer has artificial intelligence.

For the next half century, Ai has not developed much. Although the computer has the powerful memory and the data processing ability, but does not have the human cognition ability. For example, Wang, Meow Star people are silly to divide unclear. It was not until the early 21st century that deep learning appeared.

_______________________________________________________________________________________________________________ _____________________________

I have to mention a man here, Samuel (Samuel), "The Father of machine learning."

He designed a chess program that took 4 years (1956) to win a state's Checkers championship, and the era of machine learning was opened.

Processes for Input---> feature expression---> machine learning. Feature expression This part is done by manual, thankless.

Lazy people are thinking, the computer can automatically learn some characteristics, the answer is yes. Deep learning.

_______________________________________________________________________________________________________________ _____________________________

Deep learning from the achievements of a pair of Nobel laureates, their discovery is "how people understand the world."

The original signal intake of pupil intake pixels (pixel)---> Preliminary treatment of certain cells of the cerebral cortex, the discovery of the edge and direction (edges)---> abstract brain determines the shape of the immediate object (object parts)---> Further abstraction further judgment of the object (object model)

In other words, the human brain for the processing of information is layered, the idea of deep learning is to stack multiple layers, so that the output of this layer as the next layer of input, in this way, you can achieve the input information to be graded expression.

_______________________________________________________________________________________________________________ ____________

Here is the deep learning Huang Hinton

In 2006, a paper published in science opened a wave of deep learning and academia. The thesis has 2 main viewpoints;

(1) The artificial neural network of the multi-hidden layer has excellent characteristic learning ability, and the characteristics of the learning are more essential to the data characterization;

(2) The difficulty of the deep neural network in training, can be overcome by "initialization by layer"

Other words

(1) Emphasize the depth of the model structure, usually 5 layers, 6 layers, or even 10 layers of hidden layer nodes

(2) Changing the feature representation of a sample to a new feature space in the meta-space, and making it easier to classify or predict.