Nowadays, data is the core production factor of new industries and new formats, and the privacy and availability of data usage have also attracted much attention. Compared with other fields, the financial field has stricter control over data. At present, a new machine learning algorithm for data privacy protection-federated learning provides a new mechanism and platform for the application of artificial intelligence technology in the financial industry, effectively breaking the predicament of privacy protection and data use.
At present, large banks have begun to deploy, and some leading financial technology companies such as Suoxinda Holdings Co., Ltd. (abbreviated as Suoxinda, stock code: 03680.HK) have also entered the federal learning application. So what are the problems facing the large-scale implementation of federal learning in the financial industry? How to solve?
Banks “get together” federal study
“At present, some leading banks have begun to commercialize federal learning. For example, the recruitment of suppliers has begun. With the leading role of leading banks in the future, the trend of large-scale implementation of privacy computing technology in the financial industry is very obvious.” Dr. Shao Jun, a data scientist and senior researcher of Federal Learning of Suoxinda Holdings (stock code: 03680.HK) mentioned.
According to Xu Zhen, the application scenarios of federated learning in the industry are very large, which is equivalent to a data Taobao. It is expected that more federated learning-related infrastructure will be launched this year.
According to Dr. Shao Jun, a data scientist at Suoxinda Holdings and a senior researcher at Federated Learning, privacy computing is a computing theory and method for the full life cycle protection of private information. Specifically, it refers to the processing of video, audio, image, graphics, text, numerical value, and When ubiquitous network behavior information flow and other information, the private information involved is described, measured, evaluated, and fused to form a set of symbolic, formulaic, and quantitative evaluation standards for privacy calculation theories, algorithms and application technologies. Support the protection of privacy information of multiple systems integration. Federated learning is a type of privacy computing technology, a solution for machine learning and artificial intelligence to face more stringent data management regulations.
According to the “Federal Learning White Paper” released by Suoxinda Holdings, through compliant multi-dimensional federal data modeling, the effect of the risk control model can usually be improved by approximately 12% , Consumer finance enterprises and institutions have effectively saved credit review costs, and overall costs are expected to drop 5%-10% , And due to the increase and enrichment of data samples, the risk control ability is further enhanced.
The white paper mentioned that the core feature of federated learning is that all participating parties conduct joint data training without transferring their own data, so as to achieve the goal of joint modeling.
At present, federated learning has made progress in some key financial fields, such as smart risk control scenarios and privacy protection scenarios.
Dr. Shao Jun specifically stated that based on the multi-party security knowledge graph, it is possible to calculate financing service projects for small, medium and micro enterprises. Through the integration of graph computing and secure computing technology, cross-institutional (such as banks and operators) data can be realized under the conditions of protecting their respective data. Secure integration, build a joint relationship graph, break the data boundary of graph calculation, and identify more complex and comprehensive relationship chains and fraud risks.
In terms of privacy protection, it uses multi-party secure computing and image recognition technology to effectively verify the identity of merchants and cashiers, solves the management loopholes caused by equipment and manual verification in traditional acquiring management, and protects the privacy of merchants and cashiers. Effectively improve the risk control capabilities of acquiring institutions in anti-fraud and anti-money laundering on the basis of this, and reduce operating costs of acquiring institutions.
What is the difficulty of large-scale landing?
Compared with other fields, the financial field has stricter data control, so in the actual landing process, it always faces various problems.
“The realization of bridging between different privacy computing platforms has become an inevitable trend in the new situation. Different organizations apply different forms of privacy computing technology in different business scenarios; it will gradually form new data barriers that are actually difficult to connect. Therefore, searching for possible bridging agreements or even bridging platforms between different technology platforms will become a future demand.” Dr. Shao Jun from Suoxinda Holding said: “The common horizontal federated learning model is the most common and easy to achieve interconnection. Interoperability considers the joint modeling direction, but the interconnection of vertical federated learning is the main breakthrough point. Federal transfer learning itself can expand unlimited possibilities for future application scenarios.”
Regarding whether privacy computing technology can be implemented on a large scale in the financial industry in the future. Dr. Shao Jun from Suoxinda Holdings analyzed that, at present, every business line and related system in the financial field is undergoing digital transformation, and cross-departmental and even cross-institutional data circulation has become a rigid demand. As the awareness of privacy and security protection is increasing, data circulation is bound to require the support of privacy computing technology.
Speaking of difficulties in landing, Dr. Shao Jun said that there are currently a large number of small and medium-sized banks in my country, but such banks have limited technological capabilities, and have invested a lot in the preliminary construction of financial technology, and related projects have a long cycle. In addition, the market supervision of privacy computing products needs to be further improved. For privacy calculations, financial institutions should transform from passively accepting requirements to actively implementing financial-related products.
In response to the bottleneck of landing applications, Dr. Shao Jun, a data scientist at Suoxinda Holdings and a senior researcher at Federated Learning, pointed out that the first problem is efficiency. At present, large-scale general-purpose calculations cannot be done, only some simple calculations; secondly, security is due to privacy. It is difficult to quantify, and how to prove safety and how to evaluate needs to be further resolved. In addition, in addition to technology, there are more factors to consider. Such as compliance issues, how to define product attributes, and how to be recognized and supervised by relevant agencies are all issues that need to be considered.
You must log in to post a comment.