Alibaba has become the largest AI company in China. At the next day's cloud conference in 2019, Alibaba announced for the first time the scale of AI calls: more than 1 trillion calls per day, serving 1 billion people worldwide, processing 1 billion images, 1.2 million video hours, 550 million voice hours and 500 billion natural language sentences per day. In China, Ali is the only one with such huge data processing and computing capabilities.
Feitian AI platform unveiled
Jia Yangqing, president of Aliyun Intelligent Computing Platform, officially released Feitian AI Platform to the public on the same day. This is also the first machine intelligence product released by Jia after he joined Ali. According to official description, Feitian AI platform, as a structured demonstration of Ali AI's overall capabilities, will become a platform to help developers more easily master AI.
Feitian AI platform is mainly divided into high performance training and reasoning engine framework, and multi-modal algorithm development and deployment platform. According to the field introduction, through the optimization of resource scheduling ability, it can achieve 30% GPU cluster utilization enhancement, and in the large-scale sparse model training scenario, the same resource can improve the performance up to 7 times.
Industrial practice is the only criterion to test AI value. Jia Yang-qing mentioned that compared with other AI platforms, the biggest advantage of Feitian AI platform lies in its abundant practical application. It relies on such business scenarios as rookie logistics, Ali Health, Golden Map, Taobao/Tiancat, and Ant Golden Clothes, Box Horse. Feitian AI platform relies on such commercial scenarios as unmanned driving, machine translation, personalized recommendation and intelligence. It has accumulated huge experience and data in search and intelligent customer service.
The third generation of Dragon Architecture brings five times performance improvement
Ariyun's third generation of self-research Shenlong architecture also appeared on the same day. The new Shenlong architecture fully supports ECS virtual machine, bare metal, cloud native container and so on, runs through the entire IaaS computing platform, and improves the performance of IOPS,PPS and so on. Users can obtain 100% computing power on the cloud.
In addition, Aliyun also released a sixth-generation ECS instance based on Shenlong architecture, which improved computing performance by 20%, reduced memory latency by 30%, and reduced ESSD cloud disk latency by 70%. Behind the higher performance is a more inclusive price, with the highest price reduction of 58% for each product based on the new architecture.
Ali fengfu's application scenario makes aliyun a rare closed loop with both r & d and application scenarios, while developing and applying, and this dual impact approach also constitutes the unique ability of the dragon architecture to update iteratively.
Jiang Jiangwei revealed that Aliyun's millions of servers around the world will soon be fully upgraded to the third generation of Dragon architecture. Within Alibaba, the Dragon Architecture has been widely used in Taobao, Tianmao, rookie and other businesses to solve the performance bottleneck of high peak value.
Ali first disclosed the complete layout of AI
At the same time, Alibaba also disclosed the complete layout of artificial intelligence for the first time, leading the implementation of AI chip, AI cloud service, AI algorithm, AI platform and industrial AI.
In the AI chip layer, Pingtou released the world's strongest AI chip with light 800, breaking the industry record. Its performance and energy efficiency ratio is the first in the world. A light 800 is equivalent to 10 GPUs. At present, it has been applied in many scenarios of Alibaba, such as shortening the time of indexing from one hour to five minutes.
In the AI cloud service layer, the top three in the world, Ariyun, the first in Asia Pacific, has built the most complete and largest artificial intelligence cluster in Asia, including GPU,FPGA,NPU,CPU, supercomputing cluster, third generation Shenlong architecture and so on, and the public cloud service has formed the strongest support for artificial intelligence industry. A few days ago for a visual level of APP mobilization of nearly 10,000 GPU, set a world record.
In AI platform layer, Feitian AI platform, Feitian large data platform, AIoT platform and other platforms greatly reduce the threshold of AI development. Among them, Feitian AI platform is the first cloud-based commercial machine learning platform in China. It supports super-large-scale classical algorithms with tens of billions of features and hundreds of billions of training samples, reduces training costs by 35%, improves training speed by 400% under multiple scenarios, and also initiates visual modeling on public cloud, which provides developers with close to local development. The ultimate experience.
In the AI algorithm layer, Damo Institute has won the first place in the world in the fields of natural language processing, intelligent speech, visual computing and so on in the past two years. For the first time, the accurate reading rate of natural language processing in SQuAD machine reading evaluation is higher than that of human beings, and intelligent speech was selected as MIT Technology Review 2019.
Ali AI serves both internally and externally
In internal practice, Ali Xiaomi serves more than 5 million online and hotline users every day, covering 11 languages in 132 countries and regions around the world. Tmall Elves is the world's largest intelligent assistant in Chinese, with a total sales volume of more than 20 million units. Autopilot is moving from bicycle intelligence to vehicle-road co-operation. The first driverless road test license issued by Hangzhou government in 2018
Foreign service is the earliest practice of industry AI. In 2015, it took the lead in promoting the integration of Internet technology and traditional industries, and jointly cooperating with partners to promote industrial landing. Industrial AI, represented by urban brain, is being widely used in industrial scenarios. At present, 23 cities such as Hangzhou, Haikou, Beijing, Shanghai, Lhasa, Macao and Kuala Lumpur have been introduced into the urban brain.
Jin Rong, director of the Machine Intelligence Laboratory of Alibaba Damo Academy, said that Alibaba AI has formed a unique virtuous circle in technological layout and industrial landing, which truly links business opportunities with technological advantages.
Behind the shaping of Ali's strong AI strength, on the one hand, it can not be separated from the supplement of outstanding talents. Alibaba has more than 10 IEEE FELLOW,30 well-known university professors, 30 national / provincial Thousand person title experts, and has offered an annual salary of millions of US dollars to recruit young talents.
On the other hand, it can not be separated from the continuous investment of real gold and silver. Alibaba is the only company with more than half of its R&D staff among the world's leading technology companies. According to the survey data from PricewaterhouseCoopers, Alibaba's R&D investment has ranked first among Chinese listed companies for three consecutive years. Last year, Ali spent 24.7 billion yuan on R&D, which is 1.3 times that of Tencent and 1.8 times that of Baidu.