Riding the wind of rebirth

After returning to the camp from Lengshui Creek, Zhou Zhi prepared to make lunch.

Because there are almost no outsiders coming here, everyone is very confident about the safety of the camp and throws all the things here, even the three pack horses.

As a result, when Zhou Zhi had just washed the vegetables and was about to cut the meat, he heard Mai Xiaomiao's surprised cry.

"What's wrong?" Zhou Zhi ran over quickly.

"The program has finished running!" Mai Xiaomiao operated the monitoring screen on the computer: "Now we are pulling the meter to verify the correctness of the system operation."

"Whether you want to eat meat or bran, it all depends on this shop!" Li Laosan rarely said something rude.

However, when the results came out, the three were somewhat disappointed. The system analyzed the buying and selling points based on the rise and fall of the Dow Jones Stock Index from 1984 to 1995, and simulated high-frequency trading based on one million US dollars to verify its accuracy.

The system's initial design models included three categories and more than 50 types, which were reduced to 23 by Mai Xiaomiao. The main reason was that these 23 models had many excellent characteristics such as quick response, simple logic, simple learning, and accurate feedback. The rest, which were not easy to control through neural networks, were ruthlessly eliminated by Mai Xiaomiao.

Now the calculation results are out, and the final success rate is only 70%.

The three of them looked at each other in bewilderment. Although this accuracy rate allowed one million dollars to turn into one hundred million dollars in ten years, it was still completely unacceptable because those failed transactions would ruthlessly expose the existence of this system in this world, and thus arouse the vigilance of all parties.

“This shouldn’t be the case…” Mai Xiaomiao muttered, “How come it can’t even outperform the simulation algorithm?”

The algorithm model of the Baines Laboratory is a traditional data analysis model, and the success rate of the results is 90%. The Mai Xiaomiao system is as high as 95% when verifying static data, which is much higher than the data of the Baines Laboratory. Unexpectedly, after adding the time parameter to turn it into dynamic verification, it dropped so sharply.

"Could it be that the computing resources are not deployed enough?" This was the first question that came to Li Laosan's mind: "How about we expand the capacity and run it again?"

"It shouldn't be that reason." Mai Xiaomiao shook his head: "After Elbow increased Xiaozhi's computing power for the third time, we also used the gap time for running data for the Ship Research Institute. Now Xiaozhi has become a dynamic computing power allocation system, and the utilization rate is much higher, which has exceeded the 30% we expected to guarantee."

Zhou Zhi suddenly had an idea: "Xiao Miao, separate the time and count it!"

"What do you mean?"

"We divide the time into three periods and calculate the system's transaction success rate in each year!"

As Zhou Zhi spoke, Mai Xiaomiao's fingers flashed across the keyboard like dancing, and then a chart popped up on the screen.

"Yeah!" Mai Xiaomiao jumped up with joy when he saw the chart, hugged Zhou Zhi and shouted and jumped: "Success!"

"What do you mean by success?" Li Laosan looked at the two crazy people and felt that he had no sense of participation.

Zhou Zhi held Mai Xiaomiao in one hand and pointed at the computer with the other: "Our system is a learning model. In theory, the longer the data is processed and the more data is processed, the higher the accuracy will be!"

"This curve is a time-sharing diagram of the hit rate from 1984 to 1995. The hit rate in 1984 was only 50%, but in 1989 it increased to 90%, exceeding the hit rate of the Baines model. In 1994, the hit rate reached 95%. "Let's run it again!" Mai Xiaomiao said, "Use the model we have learned once to go through it again!"

"Run, run!" Li Laosan was immediately delighted and full of expectations for the system: "You scared me. I was wondering how could Xiaomiao's system have such a problem!"

This is a bit of a brainless fan. Mai Xiaomiao quickly readjusted the system and clicked Enter: "Okay, let's start the second traversal now."

This system was not deployed in Mai Xiaomiao's computer, but in the computing center of Clover Group. Mai Xiaomiao only operated the system far away in Shudu through the satellite network and Fterm in the computer.

"Still playing?" Li Laosan looked at the two of them eagerly. Now all he wanted was to get back to the grain station as soon as possible.

"Then let's close the stall." Who would still want to continue playing after it has come to this? That would be truly heartless.

It was already evening when the three returned to the grain station. Mai Xiaomiao and Li Laosan got off their horses and ran into the study to check the data results produced by the new iterative model.

"How is it?" After a long while, Zhou Zhi came in and asked the two of them.

Mai Xiaomiao pulled out a statistical table of the new data, and just as Zhou Zhi had expected, the data from the previous three years had been optimized, and the accuracy of the reach had reached more than 95%.

"We can start production now!" Li Laosan rubbed his hands ready to go.

"Take it easy." Zhou Zhi shook his head: "This data is not up to standard yet, keep practicing."

"Baines Technology said that it can go into production when the accuracy rate is 90 percent." Li Laosan said, "How come we are still not able to do it when the accuracy rate is 95 percent?"

"This brings us to the advantages and disadvantages of neuron-like systems," Zhou Zhi explained. "The power of this system lies in its self-learning function. As long as we design the model and feedback algorithm well, it can become smarter and more accurate in its judgment."

"But the flaw is that it has the same chance of having seizures as humans. When it encounters something out of the ordinary or something that has not happened during training, the results it calculates based on its own 'experience' may be very different from the actual situation we want."

"That is to say, this system may make the right judgment 100 million times, but there is a probability that it will cause us trouble 100 million times."

"What should we do then?" Li Laosan was dumbfounded.

"Brother, don't listen to my threats." Mai Xiaomiao was soft-hearted and couldn't bear to see Li Laosan getting anxious, so he quickly said, "The solution is very simple. We can add an algorithm, use our model to find the opportunity, and then use Baines' model to verify whether it is correct. Once it passes, we can trigger the automatic transaction."

"Wouldn't that mean the judgment process would be prolonged by half? That would reduce the number of trading opportunities by half?" Li Laosan was a little unhappy.

"Third brother, these are all millisecond-level transactions." Zhou Zhi smiled and said, "In 38 milliseconds, we can theoretically trigger thousands of transactions in the global market. What's wrong with a few less? We can afford less."

"Besides, compared to the fact that we want to smooth the trading curve and not let quantitative trading affect the convergence of the stock market, how can we afford this loss? What we want is safety." (End of this chapter)