当前位置：网站首页>I finished watching this video on my knees at station B

I finished watching this video on my knees at station B

2022-07-26 09:45:00 【The way of Python data】

source ： Net the world black

What kind of things need multiple references ？

What kind of things need more people to cooperate , And busy for half a year ？

Recently, there have been B The little friend who stops to brush the video should know , This is a UP Lord @ Chen zhanaotou and others are 7 month 1 Video released on May , And successfully on the hot list .

The video talks about one thing ： It took six months , stay 「 My world 」 In this game for children , Hands rub out the world's first 「 Pure red stone neural network 」, Realize the function of handwritten numeral recognition , Even the accuracy can reach 80%.

So to speak , Maybe it's too dull , There is quite a sense of seeing when I hear my family talk about the child next door who got a full score in the exam after coming home , Don't feel the charm .

Only open the barrage , Look at the full screen 「？、NB、 Oh my god 、 Witness history 」, Will understand in a trance ,UP What did the Lord do .

Anyway, when I finish reading UP After the two videos uploaded by the main , I'm on my knees , original 「 My world 」 It's really 「 The world of others 」.

Hand rub AI

If you want to be right UP I know something about this video , You have to know first 「 Redstone 」 What is this , The result from Baidu ：

Simply speaking , Redstone is 「 My world 」 The main components that transmit signals in this game , It can provide signal source for the whole circuit or part of the circuit , Signals can be passed from one part to another , You can also accept signals to respond .

According to Baidu , Redstone components are equipped with digital circuits 、 Knowledge of analog circuits , Can you in 「 My world 」 Use your imagination wantonly , Small to mechanical control ：

Big enough to rub four arithmetic units ：

Today, neural networks in the field of artificial intelligence machine learning , It is made by the wonderful characteristics of red stone , But after all, it's hand rubbing in the game , Computing power is ultimately Limited , therefore UP The master chose to appear as an introductory case 「LeNet-5」 Convolutional neural networks .

To be exact , It's a compressed castrated version 「LeNet-5」, In order to run through the model ,UP The LORD made many parts , For example, a tablet used to analog input numbers ：

Be careful , This is a single pulse pressure TABLET TABLET , It is also equipped with a display for drawing patterns in real time ：

For example, a multiplier that uses only random numbers and a single logic gate to operate decimal multiplication ：

The generation of this random number is quite interesting ,UP The main use of 「 My world 」 Settings of various tools in , Cleverly made can get 0 and 1 Unit of .

Then adjust the proportion , We get a random number that can produce different frequencies , Connect two random number generators , You get a multiplier that can calculate the product .

For example, a single neuron for receiving and output ：

Of course, there are neuron arrays ：

I rubbed these basic 「 The widget 」,UP The Lord also made many big tools , Convolution layer used to extract stroke features ：

Calculate addition through analog electricity , Then it turns into digital signal , Stacked neurons , The full connection layer used to compress and classify information ：

And activation function array ：

And the second full connection that classifies again and outputs the recognition result 、 Three layers ：

When the red stone signal runs inside 「 Hardware 」, The final output of the recognition results and confidence will be displayed ：

The overall architecture is like this ：

Is there so little that I can't understand ？ To be honest , I can't understand many places , But what you can understand , Like how to generate random numbers by detecting the stack in the thrower , It reminds me 《 Trisomy 》 Emperor Qinshihuang asked many soldiers to hold flags of different colors 01, The plot of building a humanoid computer .

Later I saw UP The introduction of the Lord , Actually this 「LeNet-5」 Model training , In fact, it's not 「 My world 」 Complete in the game , After all, it really depends on red stone alchemy , That's probably ancestral .

Instead, train the model on the computer , Deploy the final weight parameters directly to 「 My world 」 Li Lai , However, it relies on the lowest Red Stone hardware stack deployment .

The realization is ingenious and elegant , Press B The statement of the little friend standing , The silence after watching this video ...

All on their knees , You say you can't be silent ？

Who did this pioneering work UP Lord , Shortly after the video was submitted , received 「LeNet」 Architecture proposer , Turing Award winner Yang Likun's recognition forwarding .

good heavens , This wave is really out of circle .

Conclusion

To be honest ,「 My world 」 This game is not the first time for us to eat its melon , I was just addicted to 「 My world 」 On the exquisite models made by those professional architectural teams ：

After watching the video, I suddenly became more curious about this game , In especial 「 Redstone 」, So I thought about going to Zhihu to search about 「 Redstone 」 The problem of .

As it happens , I saw such a problem a few years ago —— utilize 「 My world 」 Is it possible to realize artificial intelligence with Redstone circuit ？

2019 In the year , That's what some people say ：

It's a little tricky , It's also a little crude , But there is no doubt that ,2022 Someone has been running neural network model on it for years .

I seem to understand why 「 My world 」 There are so many fans , real 、 Playable , And it has high expansibility , As long as you're serious , Romance is everywhere .

Friends who haven't seen it can go B Stand and have a look ,BV1yv4y1u7ZX, If you say 「 Oh my god 」, Remember to like and support , I've been three times in a row , It depends on you .

-------- End --------