MTL - Internet 2010-Chapter 243 data factory

After Lu Zhou left, Lu Ming returned to the laboratory.

As he walked to the back row, he instructed his assistant to remove the newly purchased graphics card and install it. The old one is the GTX280 he moved from his residence, and most of them are the latest NVIDIA GTX580 sent by the employees of WeChat.

Last year DanC.Ciresan published a paper that shocked the world. In the paper, GTX280 is used to deal with several layers of neural network. Before that, the development of neural networks has been suffering from the processing speed limit of CPU, and even if you want to use GPU, you have to make specific algorithms for specific problems.

And what makes Lu Ming feel lucky is that Lu Zhou mentioned this to him years ago and it gave him quite a headache. But just a few days ago, Dan C. Ciresan's new paper provides a fast, parameterizable convolutional neural network, which is really a sleepy pillow.

Of course, there are naturally troublesome things for Lu Ming, such as the problem of data sets, the level of interns, and the final practical application.

And all of this will take time to resolve.

After thinking about it, Lu Ming laughed. He didn't change his mentality much, but looked forward to the next job even more.

Anyway, the big things are not for him, Lu Ming, so he can study with peace of mind with his back to the younger brother.

......

Two weeks later, Mengguyun launched the crowdsourcing platform in a low-key manner.

The first reaction of most netizens who paid attention to the news was, "?!"

Although the crowdsourcing model is not new, what makes people feel fresh is the project in the crowdsourcing.

Dream Valley crowdsourcing is divided into several columns: voice-to-text, picture-to-text, picture annotation and classification, face photos and videos, foot photos, etc.

The first two items are billed based on the amount submitted, while the last two items are billed once.

Sharp-eyed Internet practitioners recognize that this Dream Valley crowdsourcing imitates Amazon crowdsourcing. It should be noted that there are a large number of datasets crowdsourced from Amazon.

Mengguyun, is this going to do something? Is it so arrogant?

Of course, it is someone else's business to guess what the outside world thinks.

a week later.

At this time, Lu Zhou was on the plane to Zheng City. Speaking of which, this was the first time he went to the city.

Next to Lu Zhou was Zhou Kai, the manager of the Guangnan branch of Menggu Promotion. Zhou Da and Wang Qiangdong behind them didn't need to come, but they were acquainted with Zhou Kai and worked together, so they also followed.

The matter is quite simple, it is nothing more than AI, and some downstream companies are needed to handle some business.

After Lu Ming's laboratory research project started, Lu Zhou first arranged for WeChat to purchase a batch of voice libraries from Haitian Ruisheng for Lu Ming's research.

Haitian Ruisheng has been in the voice annotation business since 1998. The structure of the speech database purchased by Lu Zhou can be regarded as a piece of speech corresponding to a piece of text. Such libraries are widely used to train AI, do speech recognition or conversion, etc.

As for the source of these libraries? That is naturally to manually listen and then label the text data.

With voice annotations, there are naturally pictures and videos. These are called data annotations. After a neural network is built, most of the training data that needs to be used comes from here.

The purpose of Mengguyun's online crowdsourcing or Luzhou's visit to Zheng is also here, to find people, recognize data, and practice AI.

As for the benefits, it's pretty much all-encompassing. Almost all products in Dream Valley can benefit.

"Zhou Kai."

"Boss, please speak."

Lu Zhou waved his hand, "Call me Lu Zhou. Tell me, why are you so bold that the company has just been crowdsourcing for a month, and you dare to pull up the studio to do it?"

Zhou Kai smiled and said, "That's because the boss has been rewarding me with food. As a member of the company, I have to keep an eye on the company's product status. This is not a crowdsourced process. Together they and I think it can be done, so we can arrange it directly. to engage.

We are from China in this province and know that there are many people, and the natural recruitment cost is also low. "

Lu Zhou nodded, "Then the three of you are quite strong in execution."

Zhou Kai said, "Actually, there are people doing this business in the village, and I'll be quick to get started."

After that, Lu Zhou didn't ask much, after all, it was necessary to see the specific situation before knowing the situation.

This data labeling thing is simple to say, like Zhou Kai is nothing more than finding a few who can use a computer and can start doing it. But at a deeper level, it also requires some attention.

For example, face photos and videos, or some voice metadata in WeChat, all involve some privacy and sensitivity.

If Zhou Kai can do it well and manage it properly, Lu Zhou naturally doesn't mind dividing the whole part for Zhou Kai to do it. And if it can't, then Lu Zhou also saves him from turning around and causing trouble.

Of course, it is necessary to develop some specific systems for use by annotators. It's really troublesome to think about. It is naturally possible to find outsourcing, but it is better to catch this kind of thing under its own subsidiary, and Lu Zhou will take advantage of it.

When the plane arrived in Zheng City, Zhou Kai directly took a taxi to his destination after leaving the airport.

After a while, Zhou Kai, "Arrived at President Lu. This is a park next to Xinshi."

Lu Zhou got off the car and took a look, it seemed not bad, at least not the kind of small workshop he imagined.

While walking in front of him, Zhou Kai said ~www.novelbuddy.com~ This is another introduction from fellow townspeople who are tagging. There are many companies engaged in related data business in these two office buildings. "

Lu Zhou nodded, "Like?"

Zhou Kai smiled, "I don't know if it's too much. After all, it's hard to ask if you're not familiar with it. But my fellow villager recently made a data sheet for Apple Ciri."

Lu Zhou, "Oh?"

It seems that this should be the Chinese data annotation of Apple Siri, Lu Zhou thought.

Although Apple should not have released it yet, the relevant data should be being done all the time, and the system has been training all the time.

As for how can it flow to the hometown of Zhou Kai?

That is probably only the Chinese people who can understand and recognize Chinese, and the labor cost in Zheng City is also low.

Of course, Lu Zhou also felt that what Zhou Kai did, maybe it was subcontracted layer by layer before it flowed into the park.

Although it sounds weird, Lu Zhou is more likely to come. "Data scalpers" are not normal. Doing it yourself is best to avoid it as much as possible.

Walking into the office building, the four of them got on the elevator and came to one of them.

When Zhou Kai opened the door, Lu Zhou went in and saw that it was quite normal.

The place is not big, but the office itself is in the park, and the environment is naturally better. The rows of desks are full of employees, most of whom work in front of computers.

If you really want to look at it, it looks like an ordinary company, and the scene is good.

After a while, Lu Zhou saw that a young man approached Zhou Kai for a conversation, thinking that he was a friend to work with. Lu Zhou didn't think much about it, just walked in and watched.

After walking around, although Lu Zhou was satisfied, looking at the expressionless young people in front of the computer, he felt somewhat sad in his heart.

"Maybe it's more like a factory." Lu Zhou thought.