Sha Mianzhi knew that Wang Hao also had to make a report and said that he would become his competitor, so he suddenly felt a little nervous.
He used to be Wang Hao's teacher, but Wang Hao is no longer a 'not impressed' undergraduate student. Those who can study for postgraduate or doctoral degrees at Donggang University can be regarded as elite talents, and those who can stay at Donggang University Doctoral students can be said to be the 'elite among the elite'.
Even Xu Jie, the most promising doctor under his command, who was brought here to attend the STACS conference, did not dare to say that he would be able to stay at Donggang University 100%.
Donggang University has too many talents!
Wang Hao still belongs to the very outstanding category. Leaving aside the issues in the materials laboratory, within a few months of leaving Donggang, he had completed many SCI papers and several algorithm papers, all of which were published in core journals. Among them The paper "Mathematical Model Constructed by Fourier Transform" has had a considerable impact on the Internet industry and can promote big data analysis to become faster and more accurate.
This level can be said to be top-notch.
However, because of the public opinion some time ago, Sha Mianzhi knew that Wang Hao had just completed several studies. When he first met Wang Hao, he never thought that he was here to give a report. He just felt that he was coming to the STACS conference like other scholars. It was just to gain knowledge and learn, but Ruan Hailong who was with him seemed more like he was here to give a report.
After separating from Wang Hao and Ruan Hailong, Sha Mian and his party went to the conference hall again and went to the electronic column to carefully search for Wang Hao's report.
"At the end, the second hall, the first show in the afternoon." Wang Mingkun browsed quickly and found Wang Hao's report.
"Algorithm for multiplying large numbers?"
Wang Mingkun thought about it and said, "He has just completed research on the Fourier algorithm. It is probably an improved algorithm."
He estimated, "Small results, good luck, the manuscript is over!"
"It should be like this." Sha Mianzhi thought and nodded. He had just completed a major research project. Normally, it would be impossible to complete another one quickly. The algorithm for multiplying large numbers is a focus issue, but it has not been done for decades. Breakthrough progress, as for the improvement of algorithms and improving efficiency under specific circumstances, can only be classified in the ranks of 'small research'.
"Wang Hao, I guess it was just a joke!" Sha Mianzhi immediately smiled and shook his head.
Xu Jie next to him suddenly asked, "Mr. Sha, I don't quite understand. Isn't Wang Hao a PhD in mathematics? Why does he come to attend the computer conference?"
Before Sha Mianzhi could speak, Wang Mingkun smiled and said, "Xiao Xu, if you want to go further in the direction of basic computer research, you must learn mathematics well, and it is best to learn it in depth."
"Mathematics is the foundation of computers. No matter what algorithm it is, the foundation is mathematics."
"If you are good at mathematics, it will be easy to achieve results in computer research. On the other hand, if you are only good at computers and know a lot and can use them, you can only be a programmer and cannot engage in theoretical and algorithm research."
Sha Mianzhi also smiled and said, "Listen more to what Teacher Wang said, so I asked you to read more mathematics."
Xu Jie nodded clearly.
When I think of Wang Hao, I can't help but feel a little envious. He is one year younger than me, but he has already completed influential research, can give reports at top conferences, and can talk to Sha Mianzhi and Wang Mingkun on an equal footing.
And he was just a doctoral student following his mentor.
…
The next day, the meeting officially began.
Wang Hao and Ruan Hailong arrived early. There were no special seating arrangements at the venue. They didn't sit too far forward, but just found a seat in the middle.
Not long after, the three people from Sha Mian also arrived, and they walked over and sat down side by side.
Reports in the morning were all held in Conference Hall No. 1. Only when it was the turn of some "not favored" reports would they be held in the two conference halls together.
After the opening time, the organizer of the conference, the chairman, and several committee members entered the room one after another. Then the chairman came on stage and said a lot of words, which can be simply summed up as "the meeting officially begins."
The opening of the STACS meeting was not that complicated. The chairman just said a few words, summarized last year's meeting, and then talked about the arrangements for this year's meeting, and then announced the official start of the report meeting.
The first person to appear was a young professor, probably only about thirty years old, from the Ecole Polytechnique in Paris.
The conference held by Faguo must take care of its own people, and use a study that is "not very important but can't make mistakes" as the opening remarks to get the conference officially on track.
The young professor talked about the research on 'Online Algorithm - Regularized Double Average Algorithm'. The content sounded interesting, but the research only made some progress and it was difficult to attract the attention of the scholars present.
Twenty minutes later, the report is over.
Some people in the venue applauded politely, but most of the applause came from the front row, from the organizers of the conference, judges, and invited experts. The scholars in the back row did not even give symbolic applause.
It is already polite for them not to shout "get off" or "get off" when it comes to research that is "not interesting" or "meaningless".
This is a real academic conference.
Only if you have the ability can you win applause. If you don't have the ability, just don't go up. Scholars don't understand 'hypocritical politeness'.
Wang Hao listened to the first report for a few times and found that it was of little help to him, and the research and development progress was very small. There was nothing outstanding about it. Just like everyone else, he had no interest in it.
When it came to the second report, he listened very attentively and patiently, and even used a 'teaching coin'.
Others also listened attentively.
The first report is equivalent to 'making an opening', while the second report and the third report are both more important and are research that the venue reviewers consider to be of 'significant significance'.
Many scholars who do not need to give reports come to the conference to listen to research of great significance. This top professional conference is also a place to gain knowledge and learn something. New research directions and new content can provide scholars with Knowing what others are researching and how they complete their research can help you expand your thinking and find inspiration and direction related to your own research and development.
Wang Hao values the three reports most, today's second and third sessions, and tomorrow morning's third session, which is the research between Sha Mianzhi and Wang Mingkun.
The second session currently being conducted is a study done by a professor from Oxford University, which is a theoretical study on the computational complexity of the gradient descent algorithm.
This is very rare.
Many aspects of applied research rely on an algorithm called 'gradient descent', which is a process of solving the maximum/minimum value of a certain mathematical function, from calculating the best way to produce a product to the best arrangement of worker shifts. method, the 'gradient descent' algorithm can all come in handy.
However, compared with multi-directional applications, relevant theoretical research is pitifully scarce.
The Oxford University professor who gave the report studied various aspects in a mathematical computer way from two directions: "Gradient descent algorithm does not work well on many common problems" and "Many works on gradient descent do not involve complexity theory". The intersection problem in similar situation problems is solved, thereby theoretically demonstrating the gradient descent algorithm.
Wang Hao listened with great interest. The clear logical analysis in the argument made him feel that he had a clearer grasp of the logical argument.
Another reaction is -
[Task 2, inspiration value +1. 】
After listening to the full report, I directly gained a little inspiration value from 'Task 2'. It was obvious that the gain was huge.
Although there is only a little inspiration value, you must know that 'Task 2' is to crack the number of God, and the difficulty is A level. Just adding a little inspiration value may be a very important improvement.
The Oxford University professor's report was unanimously praised, and he received unanimous applause when it was completed.
Sha Mianzhi sat next to Wang Hao and couldn't help but sigh, "It seems that it's not easy to get the best!" He was confident in his research, but it was hard to say that he should suppress the report just now, or It depends on the opinion of the conference review team.
Next is the third scene.
The second and third games on the first day of the conference can be said to be the finale. With the wonderful report just now, many people are also looking forward to the third game. The contestant is Themis Gorlik from the University of Helsinki, Finland. Si, and his colleague Almalov.
The name of the report is 'Fast and Accurate Least Mean Square Solution', and the content is about the improvement of the least mean square algorithm, which is the LMS algorithm.
After Themis Gorliks came to power, he proudly announced, “We have found the fastest and most accurate minimum mean square solution method. This method can reduce the computational complexity by more than two orders of magnitude and does not There is a loss of accuracy and improved numerical stability.”
This sentence immediately caused an uproar in the venue.
Least mean square solution is the core of many machine learning algorithms. It can reduce the computational complexity by more than two orders of magnitude. It is no joke. It is no longer an improvement, but a 'leap-forward progress'.
For example, calculating a problem requires 100 million operations, which decreases by two orders of magnitude to one million operations.
This is obviously a qualitative leap.
Themis-Gorliks began to explain seriously, and his colleague Almarov was supplementing the explanation. They proposed a very novel divide-and-conquer method, and then used the discrete Fourier transform algorithm to act as the whole The 'steering' of the architecture.
Wang Hao suddenly became energetic when he heard this.
He felt that the content of 'Steering the Helm' seemed somewhat familiar, and he would understand it if he continued to listen.
The following research content must have some reference for my own 'Fourier transform-assisted construction of mathematical model'.
"Plagiarism?"
"No, I should say application."
It is normal for the content of a published paper to be used as a reference for other research, as long as the paper contains a 'reference'.
That's okay.
However, Wang Hao couldn't help but frown as he continued to listen. He found that there were problems with the other party's research, especially involving the discrete Fourier transform algorithm. The combined 'divide and conquer method' supported the reduction of computational complexity by two orders of magnitude. Something went wrong.
The report lasted for about an hour, and Themis Gorliks finished most of the explanation. He explained the 'general direction', and also stopped to take a rest and let everyone in the venue digest it.
Everyone else was marveling at the results of the report, but Wang Hao shouted, "Mr. Gorliks!"
Themis Gorlix immediately noticed Wang Hao and asked in confusion, "What's the problem with this young gentleman?"
Everyone in the venue immediately looked over.
Wang Hao stood up and said, "Your report is very exciting, I mean the front part, but in the second part, using discrete Fourier transform to structure the overall calculation, I think there is a problem."
"The combination of discrete Fourier transform and your 'divide and conquer method', when calculating extremely large numbers or extremely large amounts of calculations, for example, more than trillions of calculations, creates a core set that bears the complexity and it is impossible to All solutions are included.”
"You rely on Caratheodory's theorem to complete the construction. The divide-and-conquer method itself has no problem, but when combined with the discrete Fourier transform algorithm, problems will arise."
"It is in the second part that characterizing the convex hull points begins..."
Wang Hao pointed in a direction.
Themis Gorlix did not look back, but said with disdain, "Young man, where is your mentor?"
"This is a STACS meeting, and you are responsible for what you say."
Wang Hao chuckled and said, "Of course I will take responsibility for my words. In addition, I am also here to give a report, not with the 'mentor'." The last sentence was said in a joke.
He continued, "And, if you heard me correctly, starting from the second part, your research mainly used my method."
