JinXing,chairman of Go AI Golaxy : Willing to provide strong assistance for Baduk industrialization

04-19-2019

Friends, distinguished guests from financial, scientific and technological circles and Baduk guests, hello.

It's a great pleasure to invite you to join us in witnessing the strategic cooperation in artificial intelligence between Golaxy Go AI and Wangxin group, which is a milestone for both sides. Wangxin group is a comprehensive financial technology service company, which can provide users with a full range of financial services. Its business covers asset management, trading platform and financial management, and has accumulated profound industry experience and industry data. In order to verify the new algorithm of artificial intelligence, we developed Golaxy. I am look forward to have the in-depth cooperation with Netcom to connect AI algorithms with industry applications, so that AI and deep learning can blossom in various fields and bring greater value.

I have a strong technical background, and my doctoral research is in the direction of artificial intelligence. After working, I have done research and development of cloud computing, virtualization, distributed system and big data. In recent years, I have returned to artificial intelligence field, focusing on deep learning algorithm.

The goal of artificial intelligence is to understand the essence of intelligence, and then create intelligent machines to replace human mental work. Deep learning is a sub field of artificial intelligence. In the last decade or so, deep learning has made a breakthrough and set off a wave of artificial intelligence. In just a few years, artificial intelligence has surpassed the highest level of human beings in image recognition, speech recognition, natural language understanding and machine translation. This series of breakthroughs is not only of great academic significance, but also of great practical significance. The industrial sector has also followed up very quickly. Many companies have actively invested in it and launched many good products to solve practical problems. 

Go originated in China, is the quintessence of Chinese culture, contains the rich connotation of Chinese civilization. Ancients said Go, who represents the philosophy of earth and heaven, has attracted countless people to study the art of Baduk. There are about 40 million fans of go in China, which is recognized as the most complex and mysterious boardgame in the world.

Go is a complicated game, but the rules is easy and straightforward . Therefore, it has became an excellent verification platform for artificial intelligence and deep learning algorithm. Many teams have joined in the exploration of go artificial intelligence, including many large teams with abundant funds and talents. With the addition of artificial intelligence, especially the strong rise of artificial intelligence go in recent years, the ancient game of go has a new brilliance. But Go is extensive and profound after all. The intelligent programs of go, including Alphago, are far away from solving the mystery of Go. There are still many unknown problems waiting for us to explore and solve.

When it comes to Go AI, we can't help saying Alphago. Alphago is the most powerful go intelligence program so far, and it is also the first artificial intelligence program to defeat human professional go players and the first world champion. The appearance of alphago has brought great shock to the world of go and science and technology.

Alphago's technology has led the revolution of artificial intelligence in go. In 2016 and 2017, alphago published two papers in nature, describing the key technologies. 4: Alphagolee, which defeated LeeSedol, mainly adopted the technology in the first paper. Alphago system is mainly composed of several parts: strategy network, given the current situation, to predict the next move; value network, given the current situation, to estimate the probability of white wins the game or black wins the game; roll out, to determine the disk situation with random simulation winning rate; Monte Carlo tree search, to connect the above three parts to form a complete system. At present, the mainstream go AI programs adopt this architecture.

The second paper introduces alphago zero. Alphago zero defeated AlphagoLee by 100:0, which is far more powerful than human beings. The key technology is reinforcement learning. Reinforcement learning uses a large number of self playing chess, and then takes the self playing chess score as the training data of the model for continuous iterative optimization. Compared with the first paper, the method of this paper is more concise, the architecture is more elegant, the implementation is easier, and the effect of improving chess power is more significant. However, this method has one drawback, that is, the computing resources consumed by self game are unprecedented. In order to build alphago zero, the Google team used 2000 TPUs for self game. Calculated by the rent of 6.5 $/ hour, the cost of the system is 310000 dollars / day. This kind of computing power consumption makes small and medium-sized teams flinch, and some teams stop the research and development of go artificial intelligence. At the same time, there are also open source projects like LeelaZero, which collect the computing power of volunteers from all over the world, hoping to complete the self game together.

For the Golaxy Go AI, it consumes a lot of computing power to verify a method that has been successfully verified, and it is not of much value. Golaxy is not a simple repetition of AlphaGoZero's paper. Golaxy learned from the basic architecture of alphago. On this basis, it has innovations in feature system, model structure, MCTS algorithm architecture and other aspects, which can be regarded as the technological evolution of Alphago architecture. It is hoped that relying on these new technologies, on the one hand, can significantly reduce the demand for training resources, and is expected to complete the model learning with less computing resources and less training samples. On the other hand, we can play different go. For example, you can play a go match that won't give in when you have an advantage, you can play games in any boards and in any komis etc. All these stuff is not mentioned the alphago architecture. Alphago has retired successfully, but the mystery of Go is far from being solved, and the innovation and exploration of Go artificial intelligence will not be stopped. We hope that the new algorithm can better migrate to other areas, with stronger practicability.

Golaxy is very happy to play with KeJie 9p in the "berry gene Cup" 2018 World AI go competition in Fuzhou on April 27. Ke jie is the best Go player China and he is a 5 times world champion. We believe that playing games is a test of mutual level, and the relationship between Go players and Go programs is harmonious in the long run. It is hoped that the two sides can put aside the winners and losers and jointly interpret the famous historical game.

With the improvement of the level of Go artificial intelligence, human-computer competition and cooperation has entered a golden period. In order to promote the development of go technology, carry forward go culture, and give full play to the greater value of high-level AI game program, Golaxy provides game analysis, territory and winrate evaluation,determine players ranks by playing with ranked Golaxy bots etc. forming a relatively complete product system. Golaxy is willing to provide strong assistance for the industrialization process of Go, so that human beings can learn, understand, appreciate and enjoy the game of Go faster, better and more deeply.

Artificial Intelligence algorithm is closely related to daily life. It is important to test the efficiency of Go training in other areas through the comprehensive demonstration of data dependence. The successful experience of algorithm engineers in commputer Go can become the general experience of artificial intelligence, such as parameter optimization method of deep learning, model optimization method, sample generation method of reinforcement learning and so on. In the process of continuous efforts to improve the level of computer Go, the deep understanding and innovation of the algorithm will be ready for the application of artificial intelligence in other directions in the future.

Wangxin group is a powerful comprehensive financial technology service group, which has set up a special artificial intelligence laboratory. From a long-term point of view, both sides attach great importance to the cooperation and strategy of the AI group. Advanced algorithm, rich data system and suitable scene are the core of this cooperation. We will not forget our original intention, strive to make greater achievements in the field of artificial intelligence and live up to your expectations. 

Thank you！

JinXing 

CEO of Shenke Technology Company,chairman of Golaxy Go AI