Recently, the “Computer Vision Youth Developer Technology and Application Conference & Developer List Awards Ceremony” was held in Shenzhen. A lot of experts and the representatives of companies from the AI field were invited to attend the conference to jointly discuss the latest technology and future trends. The deputy general manager of WeBank’s AI department, Tianjian Chen give his speech “the principle and expectation of federated visual system.
In recent years, with the rapid development in the field of AI computer vision, its application has been applied to a lot of industries, such as retail, real estate, security, medicine. In the scenario of the security checkpoint, we can use flame recognition technology in its development for airport scheduling system and people search. However, the industry is faced with problems, such as scarce labels with poor quality and scattered data. For these problems, WeBank’s AI group launch the “federated visual system” in collaboration with the Extreme Vision. It aims to solve three data issues: data soils and data lost, privacy security, regulation.
In the speech, as an example of flame recognition in the monitor camera, Tianjian Chen introduced the application scenario of the federated visual system in detail. Flame identification is for the camera to detect whether there is burning to analyze the possibility of fire occurrence and realize fire early warning. In this scenario, AI models need to collect a large number of images for training, which is affected by unstable network bandwidth and image quality. Therefore, it is difficult to increase the accuracy of the recognition model of a single company or institution. But the images of the camera may contain the company’s privacy. You cannot simply use the data shared by multiple organizations. Thus, federated learning becomes a necessity. Relying on building models locally, the federated visual system can increase the accuracy of the AI algorithm under the premise of keeping data local. In the speech, Tianjian Chen emphasized that through federated learning technology in the “federated visual system” project, the performance of the overall model was improved by 15%, which also increases the efficiency of the building model.
Federated learning in the federated visual system is essentially an encrypted and distributed machine learning technology. It allows participants to jointly build models without disclosing the underlying data and the encryption (obfuscation) patterns of the underlying data and also allows the companies to jointly build models under the premise of keeping data local. Under such a mechanism, it can successfully break down the “data soils” to realize the goal of “develop together” with the same identity and status of participants. As an initiator and heading of federated learning in China, WeBank keeps researching federated learning technology and the exploration of application implementation to promote the federated ecosystem construction in most of the industries. In addition to the field of computing vision, federated learning technology has been applied successfully to intelligent credit, intelligent risk control, price intelligence, smart retail, labor intelligence, anomaly detection and other scenarios to promote the development of intelligence city, intelligent finance and other industries.
At the end of the speech, Tianjian Chen expressed that the federated visual system can effectively help the companies expand the data application scope to share the success of data models. It also realizes that the automatic optimization of the algorithm model in terminal to further reduce the using cost of AI technology and lower the threshold of data use for companies. In the future, federated learning technology also helps more and more industries realize the intelligence of security to build a win-win federated ecosystem under the protection of data privacy.