Chapter 296: Combining theory with practice

Of course, Chang Haonan also knew that the godfather mentioned by the other three people did not actually mean the same thing as the godfather in the mafia movie.

Rather, it refers to a person who, in a religious context, is given a Christian name when he is baptized and promises to undertake his or her religious education.

The general meaning is that although the two Berkeley professors gave life to the DE algorithm, Chang Haonan and Xu Yang were the ones who gave the DE algorithm its soul.

Considering that few Chinese people would speak in this way, what the other party said should still have a certain degree of credibility.

For Chang Haonan, it is also a good thing to make a name for himself in the fields of engineering theory or algorithms that have nothing to do with confidential projects, which is equivalent to giving him an excellent disguise.

Truth and falsehood are the secret of confidentiality.

Otherwise, as a person who has done nothing on the surface, your status will rise like a rocket. It will not be difficult for anyone with a little bit of brains to figure out the reason.

At this time, the files in the floppy disk were also read.

Chang Haonan opened one of them.

"This is the algorithm from Princeton University that won the first place in last year's ICEO competition. The file below is from Kyoto University that won the second place. Since Nagoya University did not participate, they were considered last year's host. These two schools will also participate in the second place this year. It should be our main competitor."

Lin Zhi introduced it at the right time.

"First place...let me take a look."

Chang Haonan opened the first program file inside.

"The optimization efficiency of these two algorithms is indeed very high. Last month we used the optimized DE algorithm to improve the efficiency of these five test functions. Although it has been greatly improved compared to last year's third place result, I still can’t compete with them, so I thought I’d trouble you to see if there’s any way to go further.”

Suga Feian said and picked out another one from the floppy disk box:

“Here is the result of our design.”

"No rush, let's take a look at theirs first. Being so much faster than DE means they are quite capable."

In fact, Chang Haonan was a little surprised when he heard that the differential evolution algorithm only came in third place.

Even before he made improvements and optimizations, the efficiency of DE should in most cases exceed the simulated annealing algorithm that was more popular in this era and the genetic algorithm that was about to become popular on the original timeline.

Considering that the test functions we just looked at are all normal, there is nothing weird about them.

Taking third place is either because the contestant is not good at studying, or there are some other reasons——

If there really was such a powerful algorithm, it would have become famous in the academic world.

Because the competition time was limited and the test function was not complicated, Chang Haonan finished reading it in no time.

Of course, normally after reading it, the key parts need to be analyzed and tested, but with his current theoretical level, which is close to LV3, he can save the effort of the previous step to a large extent.

"this thing……"

Chang Haonan frowned and first tried to run the Michalewiczs function in the test question.

It is a relatively uncommon kind of benchmark function, at least in this era, because it is a single-objective optimization function that contains several local minima and planar regions, in which all z values ​​are almost equal. For absolute Most algorithms are not friendly.

The results are very good indeed.

"We have tried several methods, but we can't achieve their high efficiency. You can run the fifth test function again, and the results are even more exaggerated. Therefore, in the past year, some people have always suspected that these two groups were responsible for the test. Because of the fraud, the organizers only released all the relevant documents under pressure about last week."

Song Tingyu, who had not spoken just now, said in a hoarse voice.

This also explains why they insist on choosing DE even though they know that there are more efficient algorithms.

It's not because they are admirers of the DE algorithm, but because they always thought others were cheating.

But no matter what, their choice was correct.

Because Chang Haonan has already understood the little tricks played by the other party.

Compared with suspected cheating and real cheating, it is still a little bit less interesting after all.

"No need, I probably understand what they did."

As he spoke, he opened the second-ranked file and browsed it faster than before.

"what?"

Lin Zhi's surprised voice was halfway out, and she quickly covered her mouth.

They originally planned to come over today to hand the things over to Chang Haonan, and then go back and wait for news for a few days.

I was already ready to take out the gifts and leave.

The result came out on the spot?

"As long as we change the testing method, these two algorithms will be revealed."

Chang Haonan moved his body and began to build a new mathematical model.

Min f(r,n)=-n(i=1,m)R(ni)

st g1(r,n),g2(r,n),g3(r,n)

0≤ri≤1, ni∈Z+, 1≤i≤m……

Although they had no background and only had a few broad mathematical formulas, the three of them were worthy of being selected by Yan to participate in international competitions. One of them, Feian Suga, quickly saw the practical problems corresponding to this model:

"Is this supposed to be a reliability redundancy calculation for a tandem system?"

Chang Haonan has not even written annotations of the physical quantities represented by each symbol, so it can still be seen. On the one hand, it shows that his mind is really fast, and on the other hand, it also shows that he must have done a lot of homework in the past.

Physics is not pure mathematics. No matter how smart a person is, if he encounters a physics problem that he has never been exposed to before, he will not be able to see through it at a glance.

"Yes, but don't underestimate this simple series system."

Now that they understood it, Chang Haonan was happy to save himself the trouble of explanation. After running the first algorithm to start optimizing, he nodded and replied:

"There are many complex equipment, such as the over-protection system of gas turbines or internal combustion engines, which correspond to this model. How to squeeze out the potential of the equipment as much as possible and compress redundancy without affecting the safety of the system has become the optimization goal of engineers. .”

“In the past, this matter could only be accomplished with experience. Therefore, giants such as Siemens, Mitsubishi Heavy Industries, General Electric, and Ansaldo entered the industry earlier and accumulated richer data and experience. Even with the same skills, With technologically advanced hardware equipment, their unit performance indicators are also much better, which brings them more customers, which in turn provides them with more feedback data, forming a virtuous cycle. Therefore, under this model, later No one can ever surpass them."

"But with the efforts of algorithm engineers like us, this kind of control problems have gradually begun to escape the realm of pure experience in recent years. We can get some theoretical guidance to speed up the development efficiency of control systems, which also gives us those who have not caught up. It is possible for countries in the Industrial Revolution to catch up from behind.”

In fact, what Chang Haonan said is exactly what he has been doing in the past year——

Not only through his forward-looking vision, but also through technical means to improve scientific research efficiency, so that limited funds can play a greater role than others.

After hearing these words, the three people next to them were startled at the same time, and then showed different expressions, but they were all obviously thinking.

In the late 90s, the computer major in China was prosperous for a few years, but the foundation was shaky. During this period, the popularity of the accounting major was quite blind. The real first-mover bonus period was already in the early 90s. It was in the past, and the computer industry hadn't really developed yet.

As a result, these students graduated just in time for the collapse of the Internet bubble, and their encounter was almost similar to the group of people who studied biology at the beginning of the 21st century after listening to the nonsense that "the 21st century is the century of biology."

Except for a few big bosses, most people will not be able to survive the industry explosion in more than ten years, and have already changed careers.

Entering school under this background, even undergraduates from a school like Yan University may not actually know the use of what they learn. Many people are simply used to studying hard, or they pass the exam without studying hard at all. With very high scores, I chose the major with the highest score.

After all, even in the information age more than 20 years later, college entrance examination candidates’ understanding of college majors is still low enough for college entrance examination registration and career planning to become an independent industry, let alone now.

The three of them were actually just sophomore undergraduates, and had not seriously considered issues such as the application of what they had learned before.

Now when I heard Chang Haonan say this, I really felt something for the first time.

While several people were talking, the optimization algorithm running on the computer came to an end.

But unlike the previous times, it didn't really output anything.

"There is no need to continue."

Chang Haonan stopped the program and opened the log file, and quickly found the crux:

"There is a problem with the way to handle constraints, so it cannot converge, and no matter how hard I run, it will not output a valid solution."

Three faces of shock.jpg

"You see, although these two functions cannot be regarded as cheating, they actually take advantage of the competition system. The first-place algorithm is only suitable for componentized competition verification functions. It cannot be called a general optimization algorithm at all. It is almost incapable of solving complex practical problems. The ability of the second place is slightly better, but it needs to calculate the Latin square, which is not suitable for situations with too many parameters and has great limitations. In other words, the top two are test-oriented algorithms that are completely designed for competitions. .”

Chang Haonan analyzed:

"Of course, from the perspective of competition, general algorithms are not more noble than them, but personally, I still hope that the development of algorithmic science can be closer to the actual application level, and..."

When he said this, he hesitated for a moment, because the following content was indeed entirely his inference, but he finally said it:

"And based on your description, I estimate that this competition may add some restrictions on this behavior, such as adding a non-standard model or function to the test process."

"..."

In short, Chang Haonan became the last person to leave the laboratory that night.