If you have watched "Detective Chinatown", you will never imagine that in real life, a practitioner of artificial intelligence and big data will also personally participate in the real version of "Detective Chinatown", just to crack the artificial intelligence The puzzle that has landed in the industry: When training an industrial artificial intelligence algorithm model, the input data needs to be classified and labeled with industry knowledge in order to become qualified input data and used for artificial intelligence algorithm training. Such specifications and standard data.
To implement artificial intelligence in the industry, the first step is to first extract industry knowledge, and then use industry knowledge to automatically classify and label industry data, in order to truly train an artificial intelligence algorithm model that can be applied to the industry. In most industries, industry knowledge mainly exists in the minds of practitioners, so this step of refining industry knowledge must be a tedious manual method. Wu Minghui, the founder of Minglue Data, an industry artificial intelligence and big data startup, personally participated in a real-world case, just to implement artificial intelligence in the public security industry.
"My team and I were invited to Inner Mongolia to try to solve a case with artificial intelligence algorithms. Our artificial intelligence algorithm engineers brainstormed with the other police officers to convert the industry knowledge in the police officers' minds into computer-recognizable numbers. information, and then used for the classification and labeling of existing case data. However, there are still many challenges in this process, such as some cameras are out of power, resulting in a lack of critical data, or the criminals have strong anti-reconnaissance capabilities, resulting in The data is invalid. Finally, an artificial intelligence algorithm was used to automatically identify another similar case through cross-provincial concatenation of cases, and through further research and judgment, it was found that the two cases were the same criminal, and the intelligence of the two cases was combined to form a more complete chain of clues, and finally Successfully apprehended the criminal." Wu Minghui, a master of computer science from Peking University, has become addicted to solving cases.
In August 2017, Beijing Minglue Software System Co., Ltd. (referred to as Minglue Data), founded by Wu Minghui, completed the C round of financing of RMB 1 billion. The investors of this round are Huaxing New Economy Fund and Tencent Industry Win-win Fund. Founded in 2014, Minglu Data completed the A round of financing of nearly 100 million yuan in Silicon Valley Paradise Investment in July 2015; in August 2016, it obtained the investment led by Sequoia Capital China Fund, sharing investment, Ren Zixing and A round Investor Silicon Valley Paradise co-invested with a total of 200 million yuan in Series B financing. On April 18, 2018, during the 2018 Big Data Industry Summit, Wu Minghui and the author talked about the "detection record" of artificial intelligence in the industry.
The front of intelligence is "artificial"
In 2016, AlphaGo inspired the global upsurge in human intelligence; since 2017, domestic and foreign technology companies such as Alibaba Cloud, Huawei, and Microsoft have successively proposed industrial AI or industrial AI, that is, AI can be truly created by implementing AI in an industry or industry. commercial value. In addition to mentioning "big data" for the fifth time in the 2018 government work report, this year it emphasized "using new technologies, new formats, and new models to vigorously transform and upgrade traditional industries."
Artificial intelligence looks very good. Combined with industry application scenarios, the prospects look very good, and it is hailed as an engine for the conversion of old and new economic kinetic energy. However, the reality is that "intelligence" is preceded by "artificial"; if you want to apply artificial intelligence in the industry, supplementary courses cannot be avoided. This is Wu Minghui's three-step approach to the implementation of the artificial intelligence industry in early 2018, or the three steps of enterprise digital transformation: the first step is data online, the second step is data analysis and mining, and the third step is artificial intelligence. The first step of "data online" is the process of manual supplementary lessons.
As mentioned earlier, artificial intelligence algorithms need input data that meets certain specifications in order to play. A simple understanding is that data with labels is required. For example, if a picture is manually labeled as "flower", then the artificial intelligence algorithm is learning thousands or even tens of thousands of pictures of various poses labeled as "flowers". After the flower picture, it can be used to identify whether a new picture is a flower picture or a non-flower picture. In the process of labeling, one requires standardized, standardized and machine-readable industry knowledge, and the other is to manually classify and label industry data. This process of "online data" is an essential, no A short-cut process.
Minglue Data has also experienced such a painful process in the three selected industries of public security, finance, and industrial and Internet of Things. "If artificial intelligence wants to cut into the industry, it must extract industry knowledge. There are only two ways: either let industry experts learn artificial intelligence knowledge and then convert industry data into a machine-readable data format that can be understood by artificial intelligence algorithms, or let industry experts learn artificial intelligence knowledge. Artificial intelligence experts learn the industry knowledge and then do the same conversion. Compared with the two, the latter is relatively easier." Wu Minghui said that Minglue data "gnawed" the "hard bones" of the artificial intelligence industry.
Take the public security industry as an example. In 2017, Minglue Data selected different police types in more than 30 representative cities, and dispatched its own team of artificial intelligence engineers to the police force teams in these more than 30 cities. Data knowledge. "Each team has 6-10 people, stationed in a city for at least half a year, and the other party also provides certain human resources to cooperate. Through continuous brainstorming, the teams of both sides have learned the skills of all police types manually according to the needs of artificial intelligence algorithms. Data knowledge." Wu Minghui recalled this process, "Different police types in different cities have built IT systems by different technology suppliers, and data inconsistency, non-standardization, and data confusion are very common. We spent a lot of money to Do data understanding, cleaning, and integration, paving the way for the subsequent artificial intelligence algorithms."
"Data online" is undoubtedly an extremely difficult process, but once this stage is completed, the great power of artificial intelligence can be exerted later. Taking Minglue’s anti-drug big data platform as an example, traditionally, the public security is based on the technical thinking of relational databases, and builds data association libraries, element libraries, and thematic libraries around elements such as “people, places, things, things, and organizations”, which can only provide information query. , actual combat support in retrieval; Minglue platform re-labeled data and knowledge reconstruction based on artificial intelligence algorithms, and established an artificial intelligence-based data research and judgment system, especially with social communication trajectory as the entry point, combined with feature recognition And machine learning and other means to establish identification, mining and prediction models of drug-related criminals and gangs. In September 2017, the Public Security Bureau of a city on the eastern coast used the Minglue anti-drug big data analysis platform to unearth a high-risk drug-related suspect, and then expanded the network map of the suspect's relationship, and successfully unearthed a system distributed in multiple provinces. Drug gangs. This case was also selected into the "Knowledge Graph White Paper (1.0)" launched by the China Academy of Information and Communications Technology in April 2018.
"Minglue data is the initiator to help enterprises' digital transformation. The implementation of artificial intelligence in various industries requires three steps, and the core is data online. We need to really sort out the industry knowledge, and Minglue and the industry's top experts think about the nature of the industry. What is the knowledge system and how to really mark the data in the industry knowledge system. Only when this work is done well, can our artificial intelligence really land in various industries.” Wu Minghui repeatedly emphasized.
Small companies leverage industry big data
"Big data can no longer be called an industry." When talking about his views on the development of big data today, Wu Minghui believes that big data has become a social public infrastructure, which should be led by big companies such as BAT and Huawei, while startups Opportunities lie in industry applications, such as mining industry big data with artificial intelligence.
However, it is easier said than done for technology-led startups to find an effective business model for big data and artificial intelligence in the industry. IBM recently released a global executive survey report with the theme of "reverse attack of traditional enterprises". The core point is that large enterprises in traditional industries own 80% of the industry data in the society, but these data are difficult to be analyzed. Internet companies have searched and used them, so this 80% industry data is the "capital" of Internet companies that traditional companies face cross-border competition, and it is also the threshold for Internet companies to try to enter the industry.
However, 80% of the industry data held by traditional enterprises is a threshold for Internet companies and a threshold for startups. The fundamental reason why Minglue Data can obtain RMB 1 billion investment in the C round led by Tencent is that “Mininglude is committed to solving the problem of AI landing experience and bringing real value to customers. In the past four years, Minglue has been engaged in security, The three industries of finance, industry and the Internet of Things have accumulated quite a few cases, and now they have entered a stage of rapid development,” said Yao Leiwen, managing director of Tencent Investment.
The reason why Minglue Data can accumulate a considerable number of industry cases, it has carried out multi-party cooperation with industry benchmark customers such as provincial and municipal public security bureaus, Bank of Communications, ×××, China Everbright Bank, CRRC, Shanghai Metro, etc. More than 430 employees, about 75% are technical elites, the core team comes from famous universities such as Tsinghua and Peking University, and has published many papers at international academic conferences. What is more important is that Minglue Data has been down-to-earth to find the industry that can leverage the industry. Methods and paths of big data: resident scientists, project delivery teams composed of local talents, regional operation centers, flexible team incentives, etc.
After obtaining the C round of financing, Minglue Data has strengthened its investment in promoting the implementation of big data and artificial intelligence. In February 2018, former General Manager of Computer Business Department of China Great Wall Computer Shenzhen Co., Ltd., Group Sales Director and Head of Marketing Department of Aerospace Information Co., Ltd., General Manager of Aerospace Information System Engineering Co., Ltd., and Vice President of Qianfang Technology Co., Ltd. Mr. Zheng Nong, ××× Minglue Data serves as the president; former Monster WW & China Talent Network Human Resources Vice President, AMD, Nokia Human Resources Director Ms. Han Jianhong, at the same time ××× Minglue Data and serves as the vice president of human resources. Among them, Mr. Zheng Nong has been serving customers in the fields of public security, transportation, and taxation for a long time. He has rich experience in business content such as strategy, sales, and industry products, and also has a deep understanding of government needs.
In addition to increasing investment in talent, Wu Minghui also plans to use the C round of financing to increase the efforts to obtain customer resources and strengthen customer loyalty. To this end, "we will actively deepen cooperation with good IT companies in various regions." In fact, there are local IT companies everywhere, including local system integrators, software developers, solution providers, etc. These local IT companies have rich local customer resources and customer loyalty. In the past, Minglu Data cooperated with these local IT companies mainly in the form of partners, and in the future, it will also consider further strengthening the connection with end customers by means of investment.
Participating in the compilation of the "Knowledge Graph White Paper (1.0)" (April 2018) of the China Academy of Information and Communications Technology is Minglue Data's efforts to promote the popularization of knowledge graphs and enhance the industry's awareness of "data online". To put it simply, the knowledge graph is to establish a multilateral relationship between data and data to form knowledge. In the past, the formation of knowledge graphs relied on scientists to artificially summarize the laws of nature and physics, and then apply the formed knowledge structure to different data for association. Now, with the development of big data, the way that humans discover knowledge rules has been transformed into knowledge formation through automated and intelligent ways of big data analysis, mining and association. It can be considered that knowledge engineering, which studies knowledge graph algorithms, is a branch of artificial intelligence and can be used for data preprocessing and preparation of deep learning algorithms.
In March 2018, IDC China released the "China Knowledge Graph Application Market, 2018" innovator research report, which selected 5 innovative companies in China's knowledge graph market, and Minglue data was among them. According to IDC: Since 2010, there have been nearly 50 companies in the knowledge graph related market, and nearly 100 products have been launched; there are not only large Internet companies such as Baidu, Tencent, Ali, and Sogou, but also traditional solution providers such as Neusoft and Peking University. Xin, Dingfu Technology, ZTE, etc., as well as innovative companies focusing on different industries.
The reason why Minglue Data was selected into the ranks of IDC China Knowledge Graph Innovative Companies is related to the productized platform launched by Minglue Data in August last year. In August 2017, Minglue Data released the industry artificial intelligence brain - wise system. Through the AI-oriented big data management product CONA, the massive multi-source heterogeneous data is managed into industry knowledge, and then completed based on the knowledge graph database Honeycomb (NEST) The industry knowledge graph is stored, and then second-level computing and online analysis and mining are realized through the machine learning, symbolic reasoning and other technologies of the industry brain SCOPA. On the basis of this technology platform, we will gradually build a public security brain, a financial risk control brain, an industrial safety brain, etc., and finally complete the interaction between users and the platform through the enterprise-level human-computer interaction robot LiteMind.
It is precisely the way of developing a technology platform that solidifies all kinds of manual knowledge, experience and accumulation into a platform that can be run automatically, that makes it possible for Minglue Data to gradually get rid of the difficult links of manual output and data management, and gradually Embark on artificial intelligence startup work such as data collection, sorting, governance, and integration completed in an automated way. It is also based on such a technology platform. Among the nearly 100 customers who provide services, Minglue Data has achieved a 5-20% increase in the case detection rate for the public security industry, helped the financial industry to increase the risk monitoring efficiency by 3,000 times, and provided the industry with an accuracy rate. More than 98% fault diagnosis system.
Wu Minghui revealed that in 2017, Minglue Data achieved a revenue of 100 million yuan. Although it has not yet achieved profitability, Wu Minghui firmly believes that it will be worthwhile to invest in the research and development of AI in the industry. "I hope that one day, my epitaph will be written like this: The technology platform left by Minglue can create real value for several industries and truly open up the intelligent era of industry economy." This is Wu Minghui's technology platform dream , he said: "Real successful companies are patient". (Text / Ningchuan)