"If you don't want to study electrical engineering, you can get to the top of your grade, and then you want to apply for a change of major. Instead of giving up on yourself like this. Xiao Jiebao acted like he was serious, but Meng Fanqi knew very well that he was just worried that he was afraid that something like persuasion would affect his performance as a guide.

"This in itself is a contradiction, I don't want to study or can't learn a major, but I have to get to the front in this major to have the opportunity to change majors." Meng Fanqi can understand that this kind of rule is to restrict large-scale major changes, or to protect some unpopular disciplines.

You may be able to restrict him from changing majors, but will he work in this field after graduation? Is there even so many jobs in this area?

In this way, many people are trapped in the major they don't want to study, and can only change industries through graduate school entrance examinations or social training courses, so that a certain major in the university looks thriving, but in fact, it is just a cover-up.

"Besides, teacher, I didn't give up on myself, and I have been working hard to learn and research cutting-edge knowledge in new fields during the past time." Meng Fanqi pointed to his computer screen, which was a draft of his paper on "batch normalization," which focused on the problem of internal covariate shift in the training of deep neural networks.

"If I can't, I'd like to take a year off." Meng Fanqi has also considered this problem, on the one hand, he has given up the subject of electrical direction, and on the other hand, he has not yet had the capital to transfer to the corresponding major or go to other universities to continue his studies.

If you take a leave of absence, you will not be affected by some trivial matters of the school and college. But considering that it is difficult for the school not to know his parents, Meng Fanqi thought about it for a while, but still did not make this decision.

The parents at home have paid a lot for themselves, and it is difficult to come up with evidence that the second elder can understand to reassure them at the moment, Meng Fanqi does not want to affect the mood and daily life of his parents.

"Don't fool me with some English literature." In fact, the instructor already believed his words a little when he saw this, but he had already boasted to the leaders of the college this year that he would never let the student have such a thing as suspending school and persuading him to quit. "If you're still so stubborn, I'll have to ask your parents to come to school."

Meng Fanqi's eyes were cold when he heard this, if the guide really thought about him, this matter would not be unexplainable. But he knew very well in his heart that this was most likely his own little Jiujiu.

"There's no need to make such a hassle because the students' interests have changed. When we work at the university, we should pay attention to the ideas of students and the needs of students. Provide a better environment and conditions for students. What does it look like to trouble students' parents at every turn? I was kind of interested in looking at the sketches on this student's computer screen. ”

An elegant-looking middle-aged man appeared behind the instructor at some point, he was dressed a little exquisitely, wearing a pair of rimless gold-rimmed glasses, combing his back hair, and his hair was clearly defined.

"Dean Fu?" The instructor was startled when he saw the appearance of the comer, "Why did you get here?" ”

Fu Deqing is the dean of the School of Mathematics and Physics of Yenching Electric Power University, although the School of Mathematics and Physics is not a big compound in the school, but Dean Fu also serves as an important position in the school's party committee, so his status is quite high.

After the instructor who always wanted to climb up saw Dean Fu, he immediately died down and said with a smile, "Everything will be handled by Dean Fu, I won't bother, I'll go first." ”

Meng Fanqi hurriedly got up, he could see that Dean Fu was not here by chance, it was most likely that he came to find him on purpose.

But after thinking about it, I couldn't think of any place that would attract the dean of the other hospital.

"The internal covariate shift in deep neural network training will cause a serious impact on the input distribution at a deeper level, and the deep neural network has to optimize the parameters in a conservative way in order to adapt to the offset data distribution, which greatly affects the training speed and training difficulty." Dean Fu didn't explain this, but just looked at the content on Meng Fanqi's screen to himself.

"When you talk about deep neural networks, you should be referring to the hundreds or thousands of layers of neural networks you proposed, not the deep networks in the current industry, right?" Dean Fu turned his head and asked with a smile.

"Dean Fu, how are you?" Meng Fanqi was stunned for a moment, and then reacted suddenly, so far, only that girl has seen the residual network draft she prepared in advance. "That girl told you?"

"Hahaha, she's my little sister." Dean Fu smiled, "When I came here some time ago, I also came to visit me by the way, and mentioned that there is a student in our school who is studying a neural network that is hundreds of layers deep and thousands of layers, and the research is decent. ”

"To be honest, this isn't my field either, and if it weren't for my former teacher asking me to focus on it, I really wouldn't know anything. I've been looking up information for the past two days, and I said that there are hundreds of layers of neural networks, and AlexNet, which won the championship at the end of last year, is only 8 layers deep. ”

"I just had some ideas." Meng Fanqi hurriedly explained.

"Don't be nervous, I'm just a little curious to see if our school has produced a genius in this area. If so, I'm definitely in favor of it. ”

Meng Fanqi hurriedly asked Dean Fu to sit down first, and first gave him a general introduction to the background, principles and current problems.

Dean Fu himself was born in pure mathematics, and he had a little understanding of this field under the instructions of the teacher, so Meng Fanqi didn't say much, he had already mastered it.

"You think that the problem of deep networks that are difficult to train is essentially a gradient problem, so you want to use the form of residuals." Dean Fu muttered to himself, clenched his left hand into a fist, and supported his chin.

"If we don't delve into the transformation of the neural network itself, but only understand its transformation as an unknown high-dimensional function, the multiple accumulations of f(x) will be much more unstable than the residual form of x+f(x)." Meng Fanqi's later life mainly focused on understanding the process of algorithms and understanding the advantages and disadvantages of several comparison methods.

He is far less concerned with its specific inner principles, mathematical derivations, in all fairness. Now is a great opportunity to make up for this shortcoming.

"If we talk about multiple nesting of a function, f(f(f(f(.. f(x)..)))) Whether it is in the form of residuals or not, it is quite difficult to analyze. From a mathematical point of view, two nests of the same degree, with or without residuals, should not change the actual expressive power of the function. In other words, the model you build is as capable of fitting functions as it would otherwise be. ”

"If you can really see a significant improvement, then the form of this residual is just what you say the model is easier to optimize. In other words, from my point of view, the solution of the function has become more precise, and from this point of view, the residuals are a concise Eulerian forward form. ”

Fu Deqing said, picked up a pen and paper and started some deductions. Meng Fanqi was negligent in mathematical deduction, but he was familiar with the characteristics of the network built in this form, so the more he listened, the more surprised he became. Dean Fu's mathematical analysis turned out to be quite close to the results of many later experiments.

It's a pity that I watched him deduce in person, but it was a little difficult.

At this moment, as confused as him, there are several senior seniors nearby who pricked up their ears to eat melons.