Shulou(Shulou.com)06/02 Report--

Open source SQLFlow, back-feed the industry, while showing a little AI muscle.

This is Ant Financial Services Group recently the first open source to apply SQL to the AI engine project SQLFlow, the industry response.

SQLFlow, which combines difficult AI with simple SQL, greatly simplifies the threshold for data engineers to use AI technology.

It was the AI Infra team led by he Changhua, chief architect of Ant Financial Services Group's computing storage, who developed SQLFlow.

He Changhua graduated from Stanford, worked at Google headquarters for 7 years, won the company's highest technology award, and then worked at Unicorn Airbnb for 2 years, responsible for the application architecture of back-end systems.

In May 2017, he officially joined Ant Financial Services Group as chief architect of computing storage, and was selected as the 14th national Thousand talents Plan expert in 2018.

At Ant Financial Services Group, he Changhua's job is to develop a new generation of computing engines and build financial data intelligent platforms.

And SQLFlow is one of the crystallization on the main line of computing engine.

But for he Changhua, the world is changing dramatically, and he has to lead a team to explore things that no one has done.

Such as the full real-time big data intelligent system.

The cornerstone of future technology

Big data's concept first came from the search engine industry, because search engines are facing the explosive growth of huge data left by human beings on the Internet.

At the end of 2010, Google announced the launch of a new generation of search engine "Google Caffeine". The revolution of this technology is that at any time, any web page in the world can be added to the index in real time, and users can also search it in real time, solving the delay problem of traditional search engines.

He Changhua was one of the core technical leaders of the Google Caffeine development team at that time.

He explained, "the core function of Google Caffeine is real-time."

Now the goal of he Changhua working in Ant Financial Services Group is to build a "fully real-time" big data processing system, or big data intelligent platform. Because of the diversity and complexity of offline life scenarios, this is a more challenging task than building a real-time search.

He believes that this will be the cornerstone of future technology.

For computers, real-time means that the delay between sending a request and returning a response is as small as possible. For big data's processing system, it also means that the delay from data production to consumption is as low as possible, all of which means an increase in computing speed and power.

The commonly used big data computing model MapReduce, the processing of data is "piecewise", and there is a concept of boundary between slices of data. This batch processing mode will inevitably bring delay problems.

Take the search scenario as an example, if the data is processed in batches in days, it means that users will not be able to search for the updated web pages until tomorrow, and increasing the frequency of processing can partially solve the problem. Twice a day, four times a day, once every two hours.

Although it can gradually approach "quasi-real-time", the cost will also rise sharply.

In order to achieve real-time, it is necessary to break the boundaries of this batch processing, so that the data processing process is like a flow of water, along with the calculation, feedback at any time.

This also led to the later vigorous development of streaming computing engines.

In he Changhua's view, in addition to fast, "real-time system" also has two important meanings.

The first is the integration of OLTP (online transaction processing) and OLAP (online Analytical processing).

In the previous concept, OLTP has a high demand for real-time performance, while OLAP has a low demand for timeliness.

For example, using Alipay to conduct a transaction requires immediate inquiries and additions and deletions of records, which is handled by OLTP. On the other hand, the data analysis of user behavior characteristics is handled by OLAP.

But now, with the changing requirements of business scenarios, the timeliness requirements of OLAP are getting higher and higher.

For example, in the risk control scenario in Internet finance, it is necessary to judge the risk by analyzing the user's characteristic data in a very short time, which requires OLAP to be able to feedback in real time, and the feedback results can be accessed online immediately.

The second is the integration of intelligence and data systems.

Artificial intelligence and machine learning are the hottest areas of big data's application, and now the practice of most companies is to separate the data warehouse from the machine learning platform, take a batch of data from the warehouse, and put it on the machine learning platform to train the model.

With the complexity and diversification of business scenarios, this model gradually reveals problems, because whether the model can be updated in real time and whether it can be trained with more real-time data directly affects the ability to deal with complex scenarios.

"Real-time data inflows, real-time training models, real-time online decision-making and data feedback-if this line can be fully opened, it will be of immeasurable value to the business," says he Changhua.

Data, computing, intelligence, all of which constitute the "efficient big data chassis" envisioned by he Changhua, that is, a converged real-time data intelligence platform, or "Big Data Base", just as the database has become the data chassis of countless scenes.

Nowadays, there are more and more data-driven businesses not only in Ant Financial Services Group or Alibaba Group, but also in various industries.

However, the threshold of big data development is very high, if each business starts from the bottom of data development, it will be very time-consuming and labor-consuming.

How can people who do business have more energy to focus on the business?

He Changhua believes that this is not only the mission of "Big Data Base", but also the meaning of "cornerstone":

We hope to make this simple-employees in various industries and students from all business lines can easily develop upper-level applications on the basis of a solid platform without knowing the details of the lower level. How far is it from true intelligence?

Lower the threshold for data and intelligence, which is what he Changhua expects for a new engine and data intelligence platform.

At present, the financial multi-mode fusion computing engine developed by his team has realized the integration of flow computing and graph computing, flow computing and machine learning, which is getting closer and closer to his vision of "great fusion".

He Changhua revealed that the team's goal is to make the business "minimalist":

In the next two to three years, we hope that the new engine will be able to undertake the task of real-time online converged computing. Based on this engine, combined with other open source engines, we can build a whole set of data intelligent system. In this data intelligent system, the business can easily complete the process from function development to product launch, and the subsequent attracting traffic, analysis and decision-making can also be completed with the help of this platform.

He even sketched a sci-fi future scenario: you write a function to the engine, the engine will decide how many resources to use to calculate, you do not need to care about the specific calculation process, the results will be fed back to you in the shortest possible time.

When you envision a new type of business, the data intelligence platform determines what data is needed, which model to use, how to go online, and how to operate traffic.

These processes can be completed intelligently and automatically.

This is a longer-term goal. We have developed the ability to process data, and in the future, anyone can use this ability to truly democratize data.

At present, no company in the world can fully develop such a real-time data intelligent platform that combines multiple capabilities.

He Changhua also cautiously and confidently looks forward to the future: "We are also exploring, if the exploration goal is fully achieved, we will really stand in the leading position in the world."

a place where there is no one

The world is changing rapidly, and data, as a mirror of the physical world, is endless in theory. The only question is whether human beings have a way to record and collect them.

With the popularity of the Internet and mobile Internet, the cost of human behavior data collection has been greatly reduced.

With the popularity of IoT sensor equipment, a large number of data in industrial production and social life can also be precipitated.

As a result, there has been an explosive growth in the total amount of data over the past two decades.

While great digital changes have taken place in the whole world, our lives are also quietly changing.

Based on the development of data applications, we have enjoyed convenience that could not have been imagined a decade or two ago-e-commerce, O2O, mobile payment, smart home.

But in he Changhua's view, digitization is still in a very primary stage of moving offline data to online.

What we really need to think about is what kind of ability we have to deal with and apply huge amounts of data when a highly digital society comes in the future.

This is related to whether we can do more based on data, give birth to higher intelligence, and then promote the development of human society to the next stage.

This is the answer he was looking for when he returned home to join Ant Financial Services Group.

The reason for coming back is because I feel that what I am doing here, to a large extent, is an exploration for the next stage of the development of human society.

In this new exploration, dealing with huge amounts of data is a compulsory course, so he repeatedly stressed the importance of computing power: big data, artificial intelligence, deep learning. All need strong computing power, otherwise, it will be difficult to move forward.

The development trend of artificial intelligence is to simulate human ability with larger, higher and more massive computing.

"Real artificial intelligence = data + 100x computing", Google's latest artificial intelligence model level, converted to the equivalent of hundreds of GPU continuous computing for a whole year.

The new generation of computing engine and data intelligence platform developed by he Changhua and his team is actually a comprehensive carrier of efficient computing power and strong data processing capability.

It was born from Ant Financial Services Group's massive business scenarios and data, the original intention is to support Ant Financial Services Group's business, but with the gradual maturity of technology, it can also have the versatility in multiple scenarios.

Financial attributes bring high availability and high security, so that it can be widely used in other industries, not to mention dealing with life service scenarios.

The significance of this work, at large, is to promote social change, although it sounds like a grand proposition, but it is not so high above.

"every technology must have its foothold. As for Ant Financial Services Group, these technologies are closely related to the daily lives of hundreds of millions of people."

Every day, when he Changhua pulls out his mobile phone and uses Alipay to check out, he can directly feel the results of his work. Just like when he worked at Google, he used the search function every day: "the results I make, I use them every day, and I really feel that technology has changed my life."

He stated his ideal of life in this way. In the journey to the ideal, he stands at the forefront of technology and in the most daily scene, the two are inextricably linked:

Use technology to improve people's lives and promote the continuous evolution of society and people.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.