From February 7th to 12th, AAAI 2020, the top international conference in the field of artificial intelligence, was held in New York, USA. Since its establishment in 1979, AAAI has developed into one of the most watched international top conferences in the field of artificial intelligence. The conference attracted many researchers and practitioners from academia and industry to submit papers to participate in the conference. According to statistics, the conference received a total of 8,800 papers, and finally received 1,591 papers after review. The acceptance rate was only 20.6%, and the competition was extremely fierce.

Among them, the paper “FedVision: An Online Visual Object Detection Platform powered by Federated Learning” (hereinafter referred to as “thesis”), co-authored by WeBank, Nanyang Technological University in Singapore and Jiguang, a smart city solution provider in Shenzhen, was awarded “FedVision: An Online Visual Object Detection Platform powered by Federated Learning”. Artificial Intelligence Innovation Application Award”, which means that the first industrial-level application of federated learning technology in the field of computer vision has been highly recognized by the industry. It is reported that the winners of the award also include Amazon, IBM, Alibaba and so on.

The federated learning paper was recognized by the AAAI 2020 Artificial Intelligence Innovation Application Award for industrial-level applications

It is worth noting that this award for federated learning-related technologies is not the first time that it has been recognized internationally. In the field of fierce competition for AI technology, federated learning has won many awards at home and abroad, winning many important awards.

In China, the “Internet-based Crowd Intelligence Emergence Mechanism and Computation Method” project submitted by WeBank and a number of institutions was approved by the Ministry of Science and Technology of China for Innovation 2030 – “New Generation Artificial Intelligence” major project, of which federated learning is the core technology; The “Research and Application of Federated Learning Technology System” project won the 2019 “CCF Science and Technology Award” of the China Computer Federation; WeBank’s federated learning industry practice won the “Best Federated Learning Application Award” in the annual list of well-known technology media Lei Feng.com.

Internationally, Learning Federated Learning won the “Most Educational Video Award” at the top international conference IJCAI 2019; the project “Application Innovation of Federated Learning AI Privacy Protection Technology in the Visual Field” was selected into msup “2019 top100 global software” case”. Moreover, the technical application of the selected top100 cases is the relevant practice of the award-winning papers of this AAAI 2020 conference. One application, which has been recognized by the industry twice, has proved its value in both theory and practice – providing a new idea to solve the pain points in the field of computer vision applications.

For a long time, the application field of computer vision is faced with the problems of data security and privacy protection, and high transmission cost. It is not advisable to use traditional “black box” technology to capture background data to create large-scale training data sets stored centrally.

The basic principle of federated learning technology is to perform encrypted computing without the data being local, and upload the calculated model parameters to the cloud for joint modeling. With the characteristics of “data isolation, lossless peer-to-peer and mutual benefit”, each participating data “federation” can obtain a more complete model than “only based on the original independent database”, and the data is absolutely confidential. This is especially important for computer vision applications.

In addition to the elaboration of technical principles, the paper further lists practical cases of federated learning in this field.

The paper mentions that although federated learning technology has many advantages in theory, in practice there is still a lack of an easy-to-use tool to help system developers who are not experts in federated learning to integrate federated learning technology with the original system. “To this end, WeBank and Jiguang have collaborated to deploy a machine learning engineering platform to support the development of computer vision applications involved in federated learning”.

The federated learning paper was recognized by the AAAI 2020 Artificial Intelligence Innovation Application Award for industrial-level applications

Figure: Screenshot of Federated Vision Machine Learning Engineering Platform

Three large companies have already used the platform to develop computer vision-based fire risk prevention solutions and apply them to factory environments. After 4 months of deployment verification, the reliability of the scheme has been fully proved, and the feasibility of federated learning in the field of computer vision has been verified.

The award for this paper is not only a reflection of the technical value of federated learning itself, but also an important achievement of the ecological construction of federated learning. The technical research, open source tools, standard formulation, and industry implementation of federated learning have been further expanded. The application scope covers ToC, ToB, and ToG fields, and the federation ecology is becoming more and more perfect.

In 2018, the WeBank AI team submitted a proposal for a federated learning standard to the IEEE (Institute of Electronics and Electrical Engineers) Standards Association for approval, and initiated the formulation of an international standard for federated learning. Professor Yang Qiang, Chief Artificial Intelligence Officer of WeBank, serves as the chairman of the IEEE P3652.1 (Federated Learning Infrastructure and Application) standard working group. At present, the working group has held four standard working group meetings, involving more than 30 companies and institutions including WeBank, Tencent, Huawei, JD.com, and Ping An. The draft standard is expected to be released this year.

Not only that, in 2019, WeBank also open sourced the federated learning technology framework FATE (Federate AI Technology Enabler), because it can solve three industrial applications including parallel computing architecture, auditable information interaction, and clear and scalable interface Common problems, meet industrial application standards, and become the world’s first open source project for federated learning industrial applications. Since its open source, FATE has been continuously upgraded and has been equipped with the first visual federated learning tool FATE Board, federated learning modeling Pipeline scheduling and lifecycle management tool FATE Flow. Currently, FATE is included in the Linux Foundation, the world’s largest non-profit technology community.

In addition to industry technical standards and open source tools, in order to better promote industry exchanges, WeBank has written and published the world’s first monograph systematically introducing federated learning – “Federated Learning” to fully share its research in the field of federated learning. accumulation and promote industry interaction.

The book describes how to deeply combine federated learning with distributed machine learning, cryptography and security, and absorbs the principles of economics and the theory of incentive mechanism design from game theory to solve “how to ensure data security when data is not local. case, let multiple data owners share the data model” problem.

Figure: Federated Learning monograph “Federated Learning” presented at the AAAI 2020 conference

At the same time, the practicality and commercial value of federated learning have been continuously proven in practical scenarios. WeBank has used federated learning in many fields such as risk control, anti-fraud, intelligent service, marketing and retail, and achieved remarkable results. Its self-developed intelligent scoring engine is based on the vertical federated learning technology, jointly modeling the billing amount and the central bank’s credit data and other label attributes, increasing the AUC of ROC for small and micro enterprise risk control models by 12% . The “2019 China Smart Finance Development Report” led by Xiao Gang, a senior researcher at the China Finance Forty Forum (CF40) and former chairman of the China Securities Regulatory Commission, pointed out: “The application scenarios of federated learning in smart finance are very wide and there are no special restrictions. Most Common artificial intelligence algorithms such as machine learning and deep learning can be adapted to federated learning methods after a certain transformation.” The report “Artificial Intelligence Change Fraud Management” recently released by the internationally renowned consulting agency Forrester also mentioned that “federated learning is very helpful for improving the efficiency of cross-agency cooperation, and the future can be expected.”

It is believed that with the wider and deeper implementation of federated learning in the industry, the “universality” of federated learning will continue to improve, and it will have the strength to become the basis for large-scale collaboration of the next generation of artificial intelligence, and can meet the needs of technology and society and undertake artificial intelligence. The important role of intelligence in development and application.

The Links:   LTM150X0-L01 NL128102BC29-10C