In early January 2014, when it was time to bid farewell to the old and welcome the new, Meng Fanqi had many matters at hand that were fast or slow, and they all happened to be harvested at this time.
After returning from rebirth, Meng Fanqi has been unsure of how much he, a butterfly flapping its wings, can change the world.
The reason why I was able to obtain this capital and start these two companies completely relied on another time and space that people in this world cannot observe. This is quite similar to the physical meaning of "event horizon".
Moreover, "horizon" is also homophonic with "world", implying the subsequent determination to go global.
Therefore, he registered the two companies he first founded as Shijie Vision Technology Co., Ltd. and Shijie Intelligent Medical Co., Ltd. according to their technical directions.
Among them, smart medical care is still just an empty shell, and nothing is being promoted for the time being.
As for depth vision, although camera apps, related filters, functional interfaces, and functions such as payment are already being prepared intensively.
But in fact, the real core model part is still being trained.
After Meng Fanqi handled some initial matters in his company, he returned to Google's Shanghai branch again.
On the one hand, it is to complete the next two steps of the recommended advertising algorithm, and on the other hand, it is to check and accept my own FaceGAN face generation model.
"Is this considered a public weapon for private use?" Meng Fanqi really couldn't tell.
FaceGAN is a subdivision direction variant based on Meng Fanqi's previous generative network. It is mainly trained with a huge face image data set. Its purpose is to perform realistic generation and fitting of a single element of the face.
This technology currently does not have a very powerful monetization effect, firstly because the pixels it supports are far from enough, and secondly because to achieve good enough effects, the computing power required is quite large.
Only after video face-changing, AI painting, style switching of various pictures, or high-definition restoration of pictures can be done in the future will it have actual commercial value.
Therefore, this technology will be the first serious visual technology achievement that Meng Fanqi has contributed after signing a contract with Google Brain.
Although this version does not have high commercial value, the functions it implements are very symbolic and can be said to have sounded a wake-up call to human society.
On the day when Ian came up with this idea one step behind Meng Fanqi, Bengio had already predicted similar problems. In the near future, people will begin to face ethical issues related to artificial intelligence.
How to supervise, define, restrict and deal with artificial intelligence-generated content that is sufficiently fake to look real.
Humanity is still blank in this regard.
In Mountain View, California, Hinton felt relieved when he received the FaceGAN paper from Meng Fanqi.
He also heard some time ago that Meng Fanqi made money by specializing in recommending advertising algorithms at Google Shanghai, which made him very sad.
In his opinion, anyone can do these things to make money, but Meng Fanqi's talents are not used to promote artificial intelligence, especially visual intelligence. Instead, he follows the advertising team to make money.
This is simply a waste of natural resources and nonsense! It's totally outrageous.
But Google is a commercial company after all, and advertising is its main revenue product.
As an employee, Hinton found it inconvenient to express any opinions due to circumstances, so he had to give up.
Now that I saw the work of FaceGAN and learned that Meng Fanqi is still very dedicated to visual intelligence, I was immediately relieved.
"Meng! Long time no see! I'm glad that you didn't forget your visual algorithm while working on the recommended advertising algorithm!" After the video call was dialed, Hinton's expression of joy was palpable.
"I just finished reading your paper, and your ideas and operations are very clever. Change the random noise input to the generator into a low-resolution version of the face image, and let the generator do its own upsampling and fine-tuning repair."
Meng Fanqi's generative technology has been announced for more than two months, but so far no particularly decent follow-up work has been published.
What struck Hinton the most was the series of operations on human faces by the author of GAN himself, Meng Fanqi.
"The potential for generative network variants is still very large. In addition to making some fine-tuning on the structure and components this time, I also slightly modified the measurement loss function and added adversarial losses to encourage the generator to make high-resolution face to fool the discriminator.”
Meng Fanqi and Hinton had an in-depth exchange and discussion, but they were all about "generation".
Another powerful function of the GAN series network, "editing", Meng Fanqi deliberately did not mention.
This part will be the main buying point of the camera app in the next few months, and Meng Fanqi does not want to cause unnecessary problems.
Due to time constraints, he did not have that much energy to do a very detailed comparison test and publish the website.
The detailed experimental parts, as well as the [none of these people exist] face-generating website, were delegated to other colleagues on the Google Brain team.
The hidden layer editing function that has not yet been released is handed over to its own Vision Company along with the model, so that there will be an initial test version tomorrow.
"You made two more updates to the recommended advertising algorithm?" Zhou Huan saw Meng Fanqi when he came back from the holiday. Before he finished greeting him, he heard news that made his jaw drop to the ground.
"I still haven't been able to digest the exaggerated result last time." Chen Hui said a little sourly from the side. It's true that people are more likely to die than others.
Even after Christmas and New Year's Day, the two of them still couldn't recover from the previous shock. Fortunately, this kid came with two more bombshells.
"There won't be a big fuss this time, right?" Zhou Huan still remembered that after the last update was all online, although the revenue increased, due to the traffic penalty hitting too wide a range, some related customer service assistant employees They were all exhausted.
There are many unscrupulous garbage dump website members who pretend to be innocent and refuse to let go, which is very torture.
"Fortunately this time, the optimization of the mobile terminal is basically smooth and silent. The language understanding is just to give users a better search experience, and the impact will not be as big as last time."
The current search models of Google and Baidu are just temporary products and transitional products that machines cannot understand the meaning of language.
It's a bit like the telegraph before the invention of the telephone. Before the invention of the telephone, many people thought that "the telegraph was good enough and no one would use such a thing as a telephone."
But the phone is closer to people's real needs.
Search engines should also gradually become intelligent, rather than just mechanically retrieving every word entered by the user.
For example, the same keywords have completely different meanings in different sentences and contexts. This phenomenon is especially serious in Chinese.
When early Internet users searched for things, they often used a few keywords plus spaces because search engines could not understand long sentences.
If your sentence is too long and contains too much content, the search results will be cluttered, or may not be found at all.
But everything is different now.
With this prototype T method model, search engines can not only process long sentences.
And it can also handle situations where one word has completely different meanings.
Of course, it is still more difficult to deal with the kind that does not need to change a single word but has multiple meanings.
For example, "falling in love with someone" has at least four completely different meanings.
