Ma Guoning, Deputy General Manager of Taifan Technology: Mapping everything, from Königsberg to empowering all industries

On August 6th and 7th, 2022, ​​AISummit Global Artificial Intelligence Technology Conference​​ will be held as scheduled. At the "AI Empowering Industry Practice" sub-forum held on the afternoon of the 7th, Ma Guoning, deputy general manager of Taifan Technology, shared the theme of "Mapping Everything, from Königsberg to Empowering All Industries" and shared the knowledge map in detail Empowerment in thousands of industries.

If Yu Gong were AI, could he move mountains?

If Yu Gong is regarded as an AI, can he move mountains? How to move mountains?

Ma Guoning said that in the artificial intelligence industry, every vertical field is a big mountain. For example, in finance, industry, government affairs and other industries, when using algorithms to solve industry-specific problems, you will find that it is always different from the original design and implementation. The main reason is that our algorithm logic may not necessarily match the business of the industry. logic. The initial way to solve this problem is to use data intelligence and computing intelligence to train based on the amount of heap data. However, this method will also cause some bottlenecks in the later stage. To this end, we have begun to use perceptual intelligence methods such as face recognition and voice recognition to solve problems that arise in some scenarios.

When perceptual intelligence also encounters bottlenecks, the latest way is to use cognitive intelligence algorithms to solve it. So, can cognitive intelligence simulate human thinking and cognitive processes to solve complex and difficult problems?

In the field of cognitive intelligence, Google has long tried using knowledge graphs to put all knowledge into the same graph. , simulating human thinking and the process of reasoning and deduction. However, since Google proposed this idea, at least so far, there is still no way to completely simulate the human thinking process. Although the construction process of the map is not complicated, when the amount of data is large enough, various problems will be encountered. For example, WolframAlpha has more than 1 billion entities, DBpedia has more than 3 billion triples, Google currently has more than 500 million entities, and more than 10 billion relationship connections. Microsoft Probase has tens of millions of concepts alone. In this case, let alone the application, it is already difficult to just do search and query analysis.

Many scholars believe that a single point or a cluster cannot solve this problem, so they use two clusters or even a dozen clusters to solve this problem. In fact, it is difficult to perform heap clustering in the knowledge graph. The main reason is that when such a large number of entities and nodes are associated, it is difficult to separate the data.

AI empowers industries, who empowers AI?

Originally I wanted to use AI to empower industries, but from computational intelligence to perceptual intelligence and cognitive intelligence, AI now also needs someone to empower it. so what should I do now?

Ma Guoning believes that the way is to stand on the shoulders of giants.

Königsberg in the picture above is a small town, but it is very famous in the mathematics community or in the graph theory community, mainly because of a great mathematician named European Pull, solved the Königsberg Seven Bridges Problem in 1736 and created a new branch of mathematics, graph theory. When knowledge graphs are used in clusters or distributed environments, these problems need to be solved based on mathematical theory.

Therefore, when computer problems are solved to a certain extent, they will be reduced to mathematical problems. When dealing with large-scale knowledge graphs, it is necessary to divide the knowledge graph and reuse computing power to solve distributed problems. So, in the process of partitioning, how to minimize the correlation between the partitioned knowledge graphs? To this end, we need to use established or cutting-edge algorithms in the industry to satisfy graph partitioning while meeting data scale and distribution requirements.

However, currently publicly implemented or recorded algorithms cannot completely solve all problems. One is clustering and the other is distributed. Because it is difficult to simultaneously satisfy the load balancing and communication cost issues between each machine while minimizing the number of cutting edges or vertices.

How to solve these problems? Our approach is to reduce the complexity to a constant within the exponential for simple graphs without weights. On weighted graphs, reducing one of the indices to a constant complexity is a relatively cutting-edge research result. In the field of hypergraphs, the cutting problem of hypergraphs should ultimately be treated as a special case of subbrane k-part. When the K value is determined, there is no problem in solving it.

For example, for a simple graph, there are three lines of cutting. In real practice, it can be simply understood as dividing the entire knowledge graph into three clusters. Among them, the knowledge node S2 is cut independently, and S2 on the other side is a minimum independent cut. This is a simple visual description that we use to facilitate everyone to understand why this graph is separated.

In terms of effect, like the METIS algorithm, it is more balanced in minimizing the number of vertices across partitions and the time of subsequent knowledge mining; like the Hash algorithm, or the JA-BE-JA algorithm, among them On the one hand, the performance may not be satisfactory, but the performance of the METIS algorithm is relatively balanced.

Knowledge Graph and Empowerment of All Industries

Based on technology and industry research, Taifan Technology has built a knowledge graph platform. The upper layer is an application service system, including retrieval. , visual query of knowledge, intelligent question and answer, and the bottom layer builds the "vital organs" of the knowledge graph. In fact, graphs were originally a semantic problem and were developed based on the Semantic Web. The management of the semantic database, including how to update knowledge, how granular the update is, how many entities should be covered in related fields, how many mapping relationships should be covered, etc., Taifan Technology will put it into the overall framework. Therefore, this is a very versatile framework platform that can be used in all walks of life.

In addition, the entire framework also integrates necessary functions for practical applications, such as realizing the management of the entire life cycle of the knowledge base, including intelligent recommendation and retrieval. , scalability, these are all issues that must be considered in the practice of industrial implementation. In addition, many relationship exploration and mining can be solved by relying on knowledge mining.

In the following time, Ma Guoning introduced in detail the practical application of Zhimapu in various industries through scenario cases such as smart parks, smart buildings, smart transportation, smart aviation, and smart scientific data analysis.

"The starry sea of ​​technological innovation and the infinite possibilities of the future are even more exciting. I deeply believe this." Ma Guoning said that he hopes that through this sharing, more colleagues can , or other practitioners who are interested in joining this industry, can be more confident in applying artificial intelligence technology to empower thousands of industries.

​​The conference speech replay and PPT are now online. Go to the official website to view the exciting content. ​​

The above is the detailed content of Ma Guoning, Deputy General Manager of Taifan Technology: Mapping everything, from Königsberg to empowering all industries. For more information, please follow other related articles on the PHP Chinese website!