In November 8th, Tencent's global partner in Chengdu, Tencent chief operating officer Ren Yuxin put forward three strategies: connection intelligence, content ecology, technology cloud.
In the cloud part of technology, he proposed the AI in all strategy, which compared with the Baidu all in AI strategy, the difference may be changed during the period of no change. He said, AI everywhere, Tencent will all R & D technology, but not their own products and services, but open out to share with the entire industry, combined with the actual application of all walks of life entity, the value of play.
Zhang Tong, director of Tencent AI lab laboratory, comprehensively analyzes the overall strategy of Tencent AI, mainly related to the three areas of enterprises, society and academia.
Zhang Tong said, Tencent AI to build their world-class research capabilities, to do scientific and technological innovation, you need a specific research capacity. In addition, Tencent hopes to be able to AI technology transformation, application to Tencent business scenarios, but also can share with external applications research results, this is the enterprise level.
And at the social level, they want the industry to share the big ecology, from the scene, technology, talent, capital four aspects of comprehensive support AI entrepreneurs. At the same time, small and medium practitioners and traditional industries are also required to help the industry to make deep solutions to promote AI in more vertical fields.
Communication with the academic community is also indispensable, they want to cultivate talent in the combination of industry, University, industry resources, experience reverse input to AI talent
In other words, it is a combination of internal, external and Industrial Studies
In the application layer, Tencent AI mainly around Tencent internal three application scenarios, mainly in the following three aspects:
One is the game AIThe application of AI in the game, including the manufacturing process, game player experience and gaming and other. In the morning, the Tencent's chief operating officer Ren Yu Xin also mentioned the application of AI in the field of the game, he said, AI Fu can make the game world games, each character you can think out of the emotional intervention on the game. The role of the game can be self development.
Two is social AIAnd the next generation of social human-computer dialogue is an integrated hardware and software based on the Tencent, AI will launch a new product, there is great progress in the system level at present, and has been in the fall, will be open to developers next year AI.
Three is content AIAI, can be used to analyze the content, understand the content and content recommendation, Tencent AI Lab and different departments have different patterns of cooperation, some are doing a whole program, some technical points, and then provided to the business sector.
Three giant BAT, Tencent's AI capacity is still in catch-up role, it first start from the internal application scenarios, is also based on the consideration of its strategic layout, but when BT is in the application scenario of AI aimed at a broader external, can come up with the Tencent now seem to have only in the medical field
After Zhang Tong will be the Tencent AI Lab strategic architecture description again, the three application layer based on AI, including the person in charge of the game, Machine Translation and AI multimedia application scenarios and products of their respective fields in detail.
Game AI is the first plate Tencent AI application layer, artificial intelligence in the sub forum, Tencent AI lab machine learning center responsible person to Liu Han
Liu Han hopes that his share of rumor reached three points: first, what is the game AI game; two, the AI study which should have core competence; three, in order to obtain the core competence, what is the challenge, what is the solution of the Tencent.
Specifically, for the first question
Aiming at second problems
Among them, the perception ability refers to the internal state of the external environment into machine understandable; decision-making ability refers to the internal state transition for the next action and behavior; dialogue ability refers to the ability to use natural language communication.
In order to obtain these three core competencies, the game AI researchers need to face three core challenges, namely, the state space is too large, multi agent coordination, virtual and reality through. Facing the three challenges, Tencent AI Lab provides corresponding solutions respectively.
Liu Han mentioned that although the game AI is carried out in the virtual world, it has the advantages of error recognition, low cost and fast iterative speed, but it is still used in the real world of physics.
The development of multimedia AI image from AI, the latter from the shallow learning model of feature extraction and classifier training, the evolution to the end to end the training mode of the development of deep learning, it has revolutionary significance to image recognition, greatly reduced the ImageNet thousand class object recognition error rate.
With the depth of learning algorithm innovation, multimedia AI also formed, it gives the computer processing, identification, understanding of the comprehensive ability of multimedia content.
According to Liu Wei, Tencent AI explores multimedia AI from image image editing, OCR/ face recognition, image description generation, to AR and 3D vision, video shortening to some extent. At present, the promotion of the scene mainly includes film and television special effects, short video shooting, augmented reality.
In addition, Tencent AI will be in mobile applications, scene reconstruction, game entertainment, robot vision four application scenarios for promotion. Multimedia AI technology, including from the video production, processing, identification, understanding, content distribution force, so as to cover the entire video industry chain.
Machine Translation technology based on Tencent AI speech recognition, image recognition, human-computer interaction, the corresponding Machine Translation scene for simultaneous interpretation, photo translation, auxiliary translation.
According to the Machine Translation platform for Yang Yuekui introduction, difficulties in simultaneous interpretation of Machine Translation is faithfulness degree, speed of decoding problems, as well as colloquial text, punctuation, text typos Query fault-tolerant problems.
Their solution is to strengthen the attention model, reconstruct the network coverage and length of punishment; decoding, support parallel decoding on GPU, local thesaurus, priority queue and rewriting Query based on statistics, so as to achieve faster, more accurate and more general interpretation of Machine Translation.
In terms of auxiliary translation, it is understood that the auxiliary translation platform of Tencent Machine Translation platform, embedded translation system and interactive translation assistant are oriented to the manual translation team, B oriented client and C client respectively.
Yang Yuekui is also looking forward to the future of Machine Translation, in his view, Machine Translation's idea can be applied to all sequence to the sequence, including Query rewriting, summarization and question answering system, writing etc..
AI product scene
Chen Qian, general manager of Tencent Intelligent Platform Department
He believes that artificial intelligence combined with scenes have two characteristics. The first is the diversity of sensory ability, reflected in the coordination of visual and auditory information; two is the enhancement and miniaturization of computing equipment, such as smart TV such single scene single device extreme closed loop.
The scene of the principle of AI products, he summed up the four points: real, with natural and graceful speech interaction, visual and auditory information, seamless ultimate closed-loop single scene of single equipment and equipment across the scene, and to a Tencent as an example to further explain the technology to do.
Tencent AI ecological plan, the four core construction of AI ecological opening
At the forum, Tencent also released
Wang Lan, deputy general manager of Tencent open platform, said that the core of the plan is to open Tencent in the scene, technology, talent and capital resources and advantages
It is understood that, in the Tencent AI ability to open up, Tencent open platform has launched the artificial intelligence site AI.QQ.COM and Tencent AI accelerator. Among them, AI.QQ.COM has opened dozens of AI technology capabilities; Tencent AI accelerator selected 25 AI entrepreneurship projects from thousands of projects this year; in addition, there were 4 overseas AI contest winners selected. These projects cover retail, medical, translation and robotics and other 15 vertical areas of AI application exploration.
In view of this, the Tencent AI's strategic intention is very obvious, around the enterprise and industry to academia AI sharing, and then under the Tencent open strategy, the Tencent open platform will also serve as a connector Tencent and industry applications, through the gathering and open core AI capabilities and Tencent and resources, realize the scale, industrial AI technology the application of scene and landing.
* reports from the Lei Feng network