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

Since the launch of the "data Zhongtai Technology Exchange" column, it has been loved by many technical geeks. As the creator of AI-driven data center, he pays close attention to the self-research, innovation and fusion exploration of core algorithm technology, so he launches a new AI algorithm column "StartDT_AI_Lab", which mainly introduces the role and work of the algorithm team. This article first reviews the road of big data in recent years to help you understand how artificial intelligence technology has hatched from business requirements and grown into a part of the product in the process of singularity cloud development, and make a macro overview for future sharing content.

In 2013, big data became a necessary reading material for many skilled people on their desks. It is like a lamp, suddenly illuminating the complete scale of the digital world, and then people realize that the digital world has grown so big.

From then on, big data became the hottest topic in the next few years. Various concepts emerge one after another, such as "DT era", "cloud computing", "DMP" and so on. The core element is "data", and the word "data" seems to have become the norm in the technology world. Many people praise this wave of technological innovation led by big data as the "fourth industrial revolution", and compare the data to the "oil" of this industrial revolution. In a trance, there is quite a meaning that "if you get it, you live, and if you get it, you die." In the following years, companies launched a "data hoarding" movement, and those that can be stored in hard drives will never be deleted. The concept of "whether it is useful now or not, save it first, it will be useful in the future" has become the guiding ideology of CIO and CDO of various companies.

However, in the "data hoarding" movement is in the ascendant, is pushing up the price of hard disk all the way up in the development process, there is a voice like a ghost, and increasingly louder: data is useless! This voice comes from the martyrs in this movement who "unfortunately died of illness when they attacked Wei Weijie." they were once the most loyal fans of "big data," but they became increasingly confused in the pursuit of data value. finally, they were overwhelmed by the soaring costs brought about by "hoarding data" and "mining data".

Whether the data are useful or useless has become a debate topic that can be discussed. The two sides have their own opinions, and no one can convince each other, and together in this DT era, they utter such a sigh full of longing but helplessness:

If you love him, please give him data, because data is king.

If you hate him, please give him the data, because the data is dead.

So is the data useful or not? Let's take a closer look at it from the perspective of the data experience accumulated by singularity clouds over the years.

If it is just the bare data itself, it may indeed be of little use, just as a car cannot be driven by pouring crude oil.

That's why the information extracted from the data is useful, right? Not really, because most of the extracted information is fragmented and does not have direct insight into business intelligence. So the business intelligence derived from the combinatorial analysis of information is the point of value? It is not enough, because if the intelligence itself cannot be translated into decisions to implement, no amount of intelligence can bring benefits.

Is it the intelligent business of the DT era to make decisions based on intelligence and through business experience analysis? It is not complete, because the solution to the same business problem is multiple, and the decision still needs effective evaluation and optimization iteration in order to complete the data loop.

Now, a relatively complete loop of data transformation and utilization has been formed, and this is the methodology behind our mission to "make business smarter".

With this methodology as the basis, the train of thought of the overall construction will be basically clear. In the existing articles, you already know that our Simba and data Center are located in the business intelligent loop as shown in the following figure:

Obviously, this product architecture does not cover the whole business intelligence loop. In particular, starting from our vision, to help physical merchants have the same digital operation capabilities as Taobao, a very important point is to help physical merchants digitize, structure and cloud all kinds of offline non-digital and unstructured business scenes, otherwise the business intelligent loop will not be able to cover the main business scenarios of physical merchants, and the so-called improvement of business efficiency will be impossible to talk about. What is missing is the focus of this article, and it is also the main work of StartDT_AI_Lab. In order to achieve a complete business intelligent loop covering the full business scene of physical merchants, three intelligent engines are specially developed. Its position in the product architecture is as follows:

Now let's briefly introduce these three engines.

Visual intelligence engine, as a data entry, is obviously a top priority. It shoulders the important task of opening up the territory for the digital world, with the magic of turning stone into gold and turning water into oil.

When defining its product, we mainly describe it from three necessities and two general concepts:

At the strategic level:

The inevitable premise of achieving the vision of the company.

Digitize the offline scenes of physical merchants and put them online.

Offline scene in the digital transformation of the implementation cost, deployment difficulty, replication, ease of use and other aspects of the inevitable requirements.

The inevitable way in the tide of information revolution.

Expand the Internet, the Internet of everything, all online.

At the product definition level:

Whatever physical merchants need to digitize the scene.

Any computer vision technology can give full play to comparative advantages.

In terms of product function definition:

Traceable real-time multi-level tagging strategy

Basic unit-> unit attribute-> industry behavior attribute-> intelligence summary, such as artificial first-level label, wearing clothing for second-level, whether to try on a clothing for third-level, on that day, the portrait analysis of the customer for this clothing is level-four.

There are reverse requirements for business analysis and offline digitization.

Video retrieval

You can backtrack video retrieval, for example, to extract information related to secondary tags, you only need to conduct a structured upgrade and search according to the timing and location information of the first-level tags.

Vision AI + X: a specific Visual Model from the Industry

Special models such as the "double theft model"

As for the business insight engine and the business decision engine, it has been introduced in the previous column, so I won't repeat it here, just a little bit more. Big data's analysis and prediction model, which is widely used in the two engines, benefits from the structural data of the digitized offline scene provided by the visual intelligence engine, forming a series of unique big data models close to the actual business logic and experience of physical merchants.

Starting from today, this column will introduce the various types of algorithms and technologies used in their construction and the stories behind them around these three engines from the bottom up. Please look forward to it.

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.