On the fifth floor of the Shuyuan Building, Meng Fanqi and Han Ci had just come out of the dean's office.
"Let me treat you to a meal." After pressing the elevator button, Han Ci turned to look at Meng Fanqi. She felt that since she was going to trouble others, of course she couldn't let others do it in vain.
"There's no need to be so polite." Meng Fanqi knew what she meant, "If you hadn't mentioned my matter to Dean Fu at that time, Dean Fu wouldn't have noticed me in the library."
"His guidance is still very crucial to me. Part of this credit must be attributed to you." Meng Fanqi said this sincerely. If it weren't for Dean Fu's help, the quality of his early papers would have been very low. Definitely going to go down a notch or two.
"I have merit, so you can treat me to a meal. I'm hungry." Han Cike never played the trick of duplicity. Since the other party said so, she simply accepted it happily.
This guy is really rude, Meng Fanqi complained in his mind, but no matter what, it was at least very easy to get along with such a person. There will be no other ideas hidden away for you to guess all day long.
"At that time, I just thought your research in this area was interesting. When I visited Senior Brother Fu, I mentioned something about teachers and advocated paying attention to big data and artificial intelligence." Han Ci recalled the situation at that time. , "I never thought that the content would be such a breakthrough result. According to my personal opinion, residuals are indeed nothing new in mathematics and physics."
"You are a big talker. How many new things can there be under the sun?" Meng Fanqi knew that she did not mean to belittle him, and responded with a smile, "The content about the structure of the neural network itself is actually only a very small part in the direction of AI. .The residual is just a small trick that makes the network easier to optimize. From the perspective of the mathematical world, it is certainly not a big deal."
Forms similar to residuals can be seen everywhere in the mathematical world. Euler's forward method and Newton's method all contain similar ideas.
"In the final analysis, artificial intelligence is a highly practical subject. Its theory and interpretability are weak, but its performance has improved by leaps and bounds, and its industrial application is of great value. I think both the practical and theoretical schools need to objectively understand this matter Don't go back and forth all day long about things."
Meng Fanqi felt that Han Ci had begun to have a bit of a theory-oriented attitude. "I personally think that it is more important to be able to actually affect people's lives, and to be able to feel one's own achievements. Leave truly basic things like theory to geniuses like you to study carefully."
It’s not that Meng Fanqi doesn’t want to advance progress in AI theory, but that until 2023, there has been no progress in the principles and interpretability of AI. Even Academician E, after working on explaining the dynamic system perspective for a period of time and making some theoretical contributions, shifted his attention away.
"Come on." Han Ci rolled his eyes, "You say you shouldn't be contemptuous, but then you turn around and say that our research has no impact on people's lives."
"That's not to say that your research will definitely not have an impact on life, but for this subject and direction, it may be difficult to produce theoretical results. Even if they are produced, they may not be consistent with the experimental results."
Meng Fanqi does not emphasize theory over experiments, but since AI took off, there are too many unsolved mysteries that cannot be answered. The theories and laws of traditional machine learning have failed too much.
How many people have jumped into the big pit of AI interpretability, but in the end most of the explanations they came up with were far-fetched explanations, which felt like they were pointing where they wanted.
On the contrary, some practical work that focuses on visualization is very helpful in understanding the patterns of AI models.
"It was more driven by curiosity. I didn't expect that the technology you wrote about in the library would achieve a world-class breakthrough. Now that I know its value, I am really curious." Han Ci is actually the world's leader. I was the second person who knew about residual networks, but she only read it briefly and glanced at the abstract. She had no chance to read it carefully, and she didn't pay much attention to it at all.
This is normal. Who would have thought that they could meet someone in the library who has significantly broken the world's algorithm performance record?
"Speaking of which, your DreamNet paper hasn't been published yet? When is it planned to be released?" Dean Fu did not disclose Meng Fanqi's manuscript to others, but he still talked about many specific contents with his junior sister.
"Not yet. I plan to announce it when I go to Australia to attend the conference next month." Meng Fanqi was not in a hurry about this matter. He guessed that Dean Fu would not show his manuscript to others, so Only then would he specifically allow himself to answer some of Han Ci's questions.
In fact, it doesn't matter to him personally. Most of the time, the importance of articles, codes and experimental results in the field of artificial intelligence is almost the same as the theory and the article itself.
Even Meng Fanqi, who had access to the source code and conducted extensive experiments, spent more than half a year reproducing these contents. There was no risk of theft.
So Meng Fanqi readily took the manuscript from his bag to Han Ci.
"Aren't you curious why I pay so much attention to your work?" Han Ci took the paper manuscript and browsed it quickly as he walked.
"I probably want to start with the relationship between DreamNet's residuals and mathematics." Meng Fanqi doesn't pay much attention to theoretical work, but he still knows a little about some of the ideas of the Hubei academician school.
I know it, but I don’t know why. But at this point in time, it was definitely enough to fool Han Ci.
"Oh? Could you please explain it more carefully?" Han Ci was really surprised now. Dean Fu liked this junior sister very much, so he had many exchanges with her. Meng Fanqi's affairs were also mentioned.
During the exchange between the two, Han Ci felt that Meng Fanqi had a good judgment and vision on the direction of artificial intelligence, with a wide range of interests and a preference for practical technology.
His coding ability is excellent, and the theories and algorithms he proposed can be implemented with relatively high performance. However, the theoretical and mathematical definitions and deductions can only be said to be at a very ordinary level.
In the eyes of two famous mathematicians and disciples, Nature is just in the middle of the pack, with nothing particularly praiseworthy.
Han Ci fully admitted and admired Meng Fanqi's talent in artificial intelligence, but when it came to mathematics and physics, he really didn't expect him to have any unique insights.
Han Ci didn't believe it at first, so he asked him to continue talking.
"There should be many directions for residual networks from the perspective of dynamic systems." Meng Fanqi smiled secretly, and you still want to test me? I know what papers your teacher will publish in the next two years.
