Metropolis: King of Hackers!

Chapter 369 : Deep learning neural network!

Chu Hao’s living expenses in high school was earned by Chu Hao himself in Sun Feng’s computer shop. Not relying on his uncle who makes soy sauce.

In the previous life, including the cost of going to college, Chu Hao also earned it himself, and had nothing to do with the uncle who played soy sauce.

Without relatives, there is no need to celebrate New Year’s Day.

Chu Hao is alone in Jingyayuan, studying the operation of “The Brave World”. According to Chu Hao’s plan, “World of Braves” will adopt a behavioral dynamic recognition engine to play somatosensory games.

But relying solely on the way of somatosensory games will definitely cause a lot of trouble, such as how to select menus, or use items, and aiming.

If in “World of Braves”, the game character wants to use the props in the backpack, and only uses the behavior dynamic recognition engine, it is impossible for the player to get a backpack and hang it on his body, right?

But if this time, let the player use the mouse and keyboard, it will be very funny. Once “World of Braves” can’t solve this operational problem, the so-called somatosensory game is a big joke!

To this end, Chu Hao prepared a second operating system to make up for the lack of dynamic behavior recognition engine to achieve the purpose of real somatosensory operation.

The second operating system prepared by Chu Hao is

The second operating system prepared by Chu Hao is based on the calculation method of the GMM/Gaussian mixture model and is mainly used for the development of voice systems.

The so-called high-style mixed model is a model that uses Gaussian probability density function to accurately quantify transactions, and decomposes a transaction into several Gaussian probability density functions, which is a normal distribution curve.

The GMM model is usually used for speech recognition and has been widely used in the computer field. Although the speech recognition produced by the GMM model will have an error recognition rate of about 20-30%, as long as the speech is sufficiently standard and cooperates with the behavioral dynamic recognition engine to monitor the lip language, the error recognition rate will be greatly reduced.

Lip behavior dynamics in lip language, through the behavior dynamic recognition engine, can fully capture successful recognition. As for voice commands that are sufficiently standard, it is not something that Chu Hao can control. After all, the local dialects are different and can only be based on Mandarin.

Chu Hao can definitely respond. If you want to play “World of Braves” well, you must have a standard Mandarin. Perhaps with the popularity of “World of Braves”, there will be an upsurge of learning Mandarin in Xia Country!

If this is the case, “World of Braves” may be recommended by Xia Guo, and if it can be officially recommended, then “World of Braves” will definitely take off!

The GMM Gaussian Mixture Model is not complicated. In natural language processing, this is just a shallow learning neural network. In 2006, a professor at the University of Toronto in Canada published an academic paper in the top scientific journal “Science”.

Which introduces the problem of deep learning neural networks. Many hidden layer artificial neural networks have excellent feature learning capabilities, and the learned features have a more essential characterization of the data. This is the deep learning neural network corresponding to the shallow learning neural network (Chaannoy’s) network.

Before Chu Hao was reborn, in 2012, Stanford University and a large-scale computer system expert jointly used 16,000 Core.CPUs to build a model called Deep.Neural.Networks/Deep Neural Networks. Simultaneous translation from English to Mandarin, the translation process is very smooth, there is no half-point lag, wrong

The error rate is less than 1%!