Introduction: The demand for goods with trillions of people behind Double 11: Taobao creates a new generation of smart technology, Taobao becomes a super-large-scale smart APP, and cutting-edge technology reshapes the Double 11 shopping mall.
Picture: Taobao APP has become a super-large-scale smart APP
"Taobao APP has become a super-large-scale smart APP." Zhou Jingren, senior vice president of Alibaba Group, introduced on November 3, based on global The largest product cognitive map, the world's first trillion-scale cloud collaborative graph neural network per day, the scale and efficiency of intelligent computing for Double 11 in 2020 once again exceeded historical peaks.
At Alibaba’s Double 11 technical communication meeting, Alibaba CTO Cheng Li believed that the intelligent middle platform is an important part of Alibaba’s digital native business operating system, and intelligent technology flows like water in the economy. Subsequently, Alibaba announced for the first time that it had built a "cognitive intelligence engine" including the global knowledge layer, cognitive reasoning layer, and user interaction layer in its core business technology architecture.
On the eve of Double 11 this year, Taobao’s homepage was significantly revised. In user scenarios such as information flow, search, Juhuasuan, conference venues, and live broadcasts, the number of intelligent computing calls has reached hundreds of billions per day. The breadth of consumer interests has significantly expanded, various business opportunities have become more diverse, and the incubation cycle of new products is shortening. .
In addition, various basic intelligent technologies have been widely used on Taobao, with hundreds of billions of calls being made every day. In the field of visual AI, Polaritao currently supports image and video retrieval for 400 million products; technologies such as natural language learning (NLP), real-time machine translation, and semantic recognition are also key to Xiaomi, real-time translation, and product evaluation analysis. Link usage.
In the fields of information search, recommendation and marketing, traditional machine learning mechanisms will continue to fit user behavior. Cognitive intelligence can provide better discovery and richness.
Cognitive intelligence requires "real-time training" and "thousands of people and thousands of models." The computing engineering system not only faces efficiency and cost difficulties, but it can also better protect user data privacy through differential mechanisms and end-side computing. Zhou Jingren introduced that Alibaba currently uses its self-developed AIOS operating system and graph computing framework as its base, combined with a high-compression rate online learning model, to support model updates of 100 million parameters in minutes and more than 20 billion floating-point calculations in a single request.
Figure: Basic structure of product cognitive map
When Alibaba went public in 2014, deep learning and proprietary algorithms were among the core technologies disclosed to the outside world. 1. In 2016, the "Five New" plans of new retail, new manufacturing, and new energy were proposed, and the in-depth integration of intelligent technology began. The DAMO Academy, established in 2017, has currently exported more than 200 intelligent technology capabilities to business scenarios and is used daily. More than 500 billion times.
During this period, Alibaba introduced scientists from across fields around the world to carry out cutting-edge basic scientific research on automatic learning, distributed computing, multi-modal understanding, graph computing, differential privacy, etc., and actively supported technology verification in core business scenarios. , integration and diffusion. Only in cognitive intelligence related fields, it has published more than 300 top global conference papers and registered more than 180 domestic and foreign patents.