In addition to the facial recognition method implemented by Meng Fanqi when he was in China, a series of recent works in the field of medical imaging diagnostic analysis have actually been tried in California circles.

How to say that, [unanimously approved by hundreds of professors].

In particular, the two top universities in the Bay Area, Stanford and Berkeley, are two factions. Because they still insist on research when artificial intelligence is relatively declining, they also cherish these belated fruits.

Liu Yong, a professor at Stanford oncology department, was very surprised after using this series of algorithms, this computer had just learned to distinguish what a cat was a year or two ago, why was it suddenly so violent?

Especially for the more difficult diseases, a few students I brought with me can distinguish it in ten or twenty minutes, and this thing is good, and you can give an answer in a few seconds.

For this reason, he chatted with Meng Fanqi two or three times in private, and roughly understood the principle, and also provided him with a lot of data as support for coordination.

Once, Professor Liu Yong asked Meng Fanqi: "Since artificial intelligence can make such an accurate analysis and judgment of the type, area and contour of the lesion in the picture, can you help write the doctor's order or text analysis?" ”

Meng Fanqi was speechless after hearing this, but he didn't expect Professor Liu to accept new things quite quickly, and he was already daydreaming.

He had to tell Professor Liu the truth, not to mention the multimodality of the combination of language and images, the language model itself is still a relatively urgent breakthrough field.

I'm afraid it will be a few years before he realizes the functions he needs.

Other medical professors at Stanford, such as Director Jeffrey, are very optimistic about Meng's Alpha Fold project, especially Jeffrey, who has served as the principal investigator of more than two dozen clinical drug studies, so he is very well able to understand the value of this protein analysis capability.

Jeffrey was a brainy, somewhat chubby middle-aged man with a red face and a warm attitude towards his students.

After learning that Meng Fanqi's Alpha Fold project lacked high-quality protein data, he was also very active in providing assistance.

Overall, the Stanford School of Medicine is enthusiastic about the breakthroughs of its students, and with the influence of these two top universities in California, Meng's achievements are slowly radiating outward.

It's just that the preparation of a large amount of data is not something that can achieve significant results in just a few days.

Even with the collective support of relevant professors in California, the speed of data accumulation is still much slower than Meng Fanqi thought.

The official opening of the Alpha Fold project will be postponed by at least one to two months.

Therefore, in the period before the start of school, Meng Fanqi's protein analysis plan had to enter a period of stagnation.

"If the technology is advanced too quickly, you will encounter this kind of problem." Meng Fanqi was a little helpless, not to mention that many fields had not had time to accept and digest the ability improvements he had made.

The data that was originally accumulated in these places is completely inadequate in the face of new technology, and the amount is too small.

Even if they can turn the head of the car in time and start accumulating and annotating high-quality data, it will definitely take a while.

The data can't keep up, and even if Meng Fanqi's theory is good, it is difficult to make a good enough effect to be convincing.

"I've been working on application technology for a good month, and now I have to wait a month or two for the data, and it seems that it's time to do some basic work to pave the way for the future."

After many months, Meng Fanqi was finally forced to calm down and have enough time to do some basic theoretical methods, instead of rushing to make products or realize them.

After all, whether it is AI-generated speech, images or text, there are still many flaws and problems in the current technical theories.

The most recent thing he plans to develop is a very popular image generation technology, the basic component of Stable Diffusion, the diffusion principle.

This is the cornerstone of many of the later good build techniques, and it's perfect for preparing now.

The diffusion model, a relatively unclear term, has been heard many times, although few people have known the principle since then.

From the works generated by AI drawing software to beat a group of human artists and win the championship in the digital art category, to Midjourney, Imagen, novelai and other domestic and foreign platforms have blossomed everywhere.

More and more people have clicked on relevant websites to try to let the AI paint the picture in their minds, or make partial modifications.

There are those who summon the ancient gods with various mysterious spells, and there are also those who use the pictures to make all kinds of magical jokes.

In 2022, AI painting, AI generated images, advanced several times in just a few months.

Every breakthrough has brought about a visible improvement, far beyond human imagination.

At the end of the 22nd year, everyone was still laughing at what AI drawing was, it was too ugly.

As a result, after three months, it turned out that things were not so simple, and the AI began to produce a large number of various choppy plots, which attracted the attention of a considerable part of the people.

At that time, many people were still joking that although the AI couldn't draw, don't say it, it still has a very good grasp of the subject matter! The level is not enough, and the subject matter is to make up.

After another three or four months, by the end of '22, the level and ability of AI painting will no longer be disputed.

This time, the main point of contention has changed again, and it has become whether AI drawing is plagiarism or not, and who is stronger in AI level or artist.

Putting aside the unclear question of who is stronger, it can be seen from the content of everyone's argument that the ability of AI painting has indeed improved very quickly.

"In all fairness, when it comes to AI generation, the GAN generative approach is a detour."

Although Meng Fanqi's GAN generative method has been widely praised in the academic community and has established a high academic reputation and popularity, and the effect of FaceGAN's fake face generation is also quite amazing, it is the diffusion model that really makes AI mapping popular.

"The adversarial mode of the GAN method is certainly amazing, but it is still very troublesome for two networks to learn from each other." Meng Fanqi thought for a moment, the big problem now, data is one thing, and computing equipment is another.

Originally, I released the technology in advance, and the graphics card was already very insufficient. At present, it is still possible to use GAN to target a specific thing, such as FaceGAN, which only makes faces, but it is more difficult to generate it directly from text.

The principle of the diffusion model is actually not difficult, mainly by adding noise to the photo, and then learning the various characteristics of the current image in the process. After that, a random noise image that obeys the Gaussian distribution is generated, and then the noise is reduced step by step until the expected image is generated.

The code is not so difficult to write, but if it is written as a paper and pondering the principles, the mathematical logic and derivation in it are enough for Meng Fanqi to drink a pot.

"In terms of mathematics.... In the past two days, let's ask Han Ci and Dean Fu to help, I haven't contacted them much for a long time. Or you can ask Hinton and Li Feifei for some pointers, both of whom are my mentors. "When it comes to math problems, it's natural to be reassuring to find math professionals.

Although Li Feifei and Hinton are not from mathematics, Li Feifei has a physics background, and Hinton is the godfather of the field, and they must be very strong in this regard.

After creating a new folder, Meng Fanqi was just about to start working, but he vaguely felt that something was wrong, and he seemed to have forgotten something.

It took me a while to remember that my hapless roommate Don Juan was coming to the Stanford neighborhood today.

"After more than half a year, I almost forgot that I was still a student."

Meng Fanqi, who has been intensively developing for three or four months, feels as if more than a year has passed.

Don Juan is here, so let's give myself a few days off, I've been here for a few weeks, but I haven't been able to visit the Stanford campus properly.