Chapter 319 Neural Network

Just when Lynch was addicted to and hesitated about the interface module of the processor built by this evil spirit with the power of fate, he suddenly realized that the chip that was also starting to work in his mind was heading towards a new chapter.

It is very different from the simple CPU chip model he once conceived, nor is it the GPU chip model that has changed its mind recently.

It's a more extreme chip model.

AI chips.

As we all know, chips themselves have many types. If they are divided by process, microcomputers and mobile phones are the keys to consumer electronics, and they are naturally assigned the best consumer-grade chips. For example, the latest flagship phones launched by various mobile phone manufacturers every year, if they are not matched with the latest 870/880 chips, they are absolutely sorry for the title.

Even if some chips here have severe heat or even huge costs due to their performance, the latest and most powerful name cannot be underestimated, otherwise consumers will teach you how to behave this year.

In addition to these chips, there are different types of chips left. They do not need to use the most advanced 5nm process. They can even be controlled by microcontrollers, including ARM, DSP, etc., which are general terms mcu chips.

No matter how high their manufacturing process is, it is only 28nm level, but it is extremely costly. For example, cars are the consumption of such chips. The simplest two windows are required to control lifting and lowering, not to mention complex functional modules such as automatic assisted driving.

AI chips are a chip type that is more extreme than GPUs in terms of alienation.

If it is a GPU, it is an ALU unit (arithmic logic unit) that requires more than CPU.

Then AI chips are special chips customized for AI algorithms, so they consume less energy and have higher efficiency when executing AI algorithms.

Lynch initially saw the "processor" module that inspired him to create efficiency with the "Holy Word of Creation", and soon discovered the difference between it and the ordinary chip structure.

For example, in the category of autonomous driving, ordinary CPU processor calculations, because computing is not a strong point, the speed cannot meet the needs. As for the GPU chip, it is still satisfactory, but its excessive cost and power consumption often exceed the consumer's tolerance range.

At this time, AI chips specially customized to suit these application scenarios came into being, such as the graphics card chips that Google used to train AlphaGo in the early stage, and later it directly used self-developed AI chips to train.

Only then did Lynch remember it in a daze.

The reason why AI chips can win is that AI algorithms involve too many convolutional, residual network, and full-connection type calculations.

And these calculations are essentially addition and multiplication.

Similar to the calculations of those spell models that Lynch once came into contact with.

You should know that if a more mature AI algorithm executes it once, it is often equivalent to trillions of addition and multiplication calculations.

For more advanced CPU processors, the number of calculations in one second is only tens of billions of times.

There is a time gap when dealing with it for trillions of times.

But like the TPU1 developed by Google, its calculation times in one second is close to 100 trillion times.

The AI ​​algorithm that has trillions of calculations has been executed hundreds of times in a second.

If the GPU is specifically separated from the CPU to process image calculations, then the AI ​​chip is specifically separated from the CPU to process AI algorithm calculations.

All of this comes from the dependence of deep learning on neural network algorithms!

But it happened.

At this time, Lynch looked at the miracle that had been built in his mind and could not say a word.

To say so much, it is redundant.

The spell model itself involves the most basic addition and multiplication operations.

The plan that Lynch originally formulated was to transform into artificial intelligence in the future, but unexpectedly, he was forced to take a step forward here.

At this time, he looked at the gods opposite him again, and the other party looked at Lynch with satisfaction.

Obviously, Lynch understood the structure of AI chips and would not let the orbs beads be covered in dust for a day.

"Neural Network!"

The evil spirit made a shocking sound, sweeping across Lynch's eardrums again.

And in his mind, all the information about this algorithm was suddenly reorganized, and combined with the part that was rewarded by knowledge.

Something that is indescribable was first dug out, and when it came out, it was a CPU that was specially customized mining machines, and these mining machines used AI chips.

In the field of computing, ACIS (AI chip) has emerged from the big encirclement between CPU and GPU.

Lynch curled his lips.

Spell.

magic.

Spell model.

In terms of the most reliable spell casting, it is naturally to teach the processor to complete the entire spell casting process by itself.

The external PID handles the overall secret energy field parameter problem, while the internal one is the calculation problem of the AI ​​chip processing spell model.

people.

It shouldn't exist in this link at all.

Letting the chip learn to use magic is only the first step.

The second step is to let the chips learn to make choices!

Human reaction has been proven that it cannot be less than 0.1 second, so sprinting believes that the reaction speed exceeds this is a rush.

However, in the face of the ever-changing spell battle, if Lynch thinks of a 1V1 single fight, then it is enough to rely on himself.

But if he wants to become a ten thousand enemy in spell warfare, he also needs an automatic spell response mechanism.

This is also the reason why countless mages need to specify magical plans for the next battle in advance, because their thinking can no longer support millisecond response battles, and can only make a more comprehensive plan and then embed it into their instinct.

Since he remembers that there is an AI chip that is about to be born inside the palace, why not go straight and develop the magic response as well?

And here we have to go back to the original question.

If a machine handles 1+1, it can crush everything in the world.

But it takes a long way to know how to choose spells by machines!

It takes countless manufacturers' efforts to let machines replace humans to drive, and they are still hovering in L2 today.

What is machine learning?

To put it simply-

Person:1+1=?

Machine: 5

Person: 1+2=?

Machine: 7

Person: 3+2=?

Machine: 10

After countless times...

Person:1+1=?

Machine: 2.

The so-called artificial intelligence.

The more artificial there is, the more intelligence there is.

There was once an example of a mango away.

For example, if you want to choose mangoes, but don’t know what kind of mangoes are delicious, you need to taste all the mangoes first, and then summarize the deliciousness of dark yellow. Then buy them and choose dark yellow.

Machine learning is to let the machine taste all the mangoes first and let the machine summarize a set of rules.

What the people here need is to describe the characteristics of each mango to the machine, from color to soft and hard, and finally let it enter whether it tastes good or not.

The rest waits for machine learning to find out a set of rules to judge that the "dark yellow" mango is delicious.

This learning process is machine learning, and neural networks are the most popular machine learning method.

Lynch calmed down again, walked to the bookshelf of the Palace of Memory, and quietly opened the original books.

The progress jumped too quickly, which forced him to work overtime and learn about the rest of his knowledge. He was like a chef who had just started to turn on the recipe.

Although the situation is a bit urgent, it is also destined.

The algorithms used by AlphaGo were Monte Carlo algorithms and neural network algorithms, and neural network learning is an inescapable barrier for all those who engage in machine learning.

This is also a knowledge point that Lynch needs to quickly chew off.

At this time, he was sitting in a cage, deducing it on the muddy ground with nothing else in his heart, without any concern about the filth and sand on it, as if this was a wide-screen blackboard for him to calculate.

Neural networks, as the name suggests, come from human neurons.

Basically, through biology teaching in high school, most people can understand the principles of neurons. There is a spherical cell body in the middle and a small and prosperous branch of nerve fibers at one end, the scientific name of dendrites.

On the other end is a long protruding fiber, scientifically known as axon.

The function of neurons is that each dendrites receive electrochemical signals from other neuronal cells. After these pulses are superimposed on each other, once the final intensity reaches the critical value, the neuronal cell will be activated and then send a signal to the axon.

The axon changes the membrane potential through the exchange of nanopotassium ion inside and outside the cell membrane, so that the entire electrical signal is transmitted without decay.

Finally, these signals are transmitted to other axons and dendrites, which stimulate them to generate signals and become secondary neurons.
To be continued...