Wang Cong: Cutting-edge research on algorithms is committed to reducing data bias and improving privacy protection

　　Zhongxin Jingwei Client, July 19 (Zhang Yanzheng) Recently, the seminar on "Intelligent Algorithms and New Formats of Digital Economy" sponsored by the Digital Economy Research Center of Chinese University and the Interdisciplinary Platform of Digital Economy of Chinese Min University was held online. Wang Cong, an assistant professor at Peking University's Guanghua School of Management, said at the seminar that cutting-edge research on algorithms is committed to reducing data bias and improving privacy protection.

　　Wang Cong pointed out that at present, algorithms are widely used in e-commerce, financial technology, smart medical and other fields. For example, in terms of e-commerce, the recommendation system is based on the consumer's massive purchase history data, extracts the preference characteristics of consumers to purchase goods, and then makes personalized recommendations, thereby saving consumers' information search costs and facilitating them to make simple and efficient decisions. For merchants, increase your product conversion rate so they can get more money.

　　"In terms of financial technology, whether it is investment decisions or the issuance of consumer credit, it is inseparable from algorithm support." Wang Cong cited Alipay, Huabei, and JD.com As an example, on similar online consumer finance platforms, individuals provide very little information, and the platform can decide whether to allow consumers to use related financial service products. In fact, the platform has used the information accumulated by customers to dock to the new business.

　　Wang Cong also said that high-quality medical care is a highly scarce resource, and the distribution is uneven, and intelligent medical care can be eliminated through algorithms, which can eliminate the difference in medical level. "If the diagnosis and treatment records of high-quality doctors are combined with the patient's physical signs information, they are jointly constructed into a training set of algorithms, and then a diagnosis and treatment plan is made, which can be used by less developed areas to make auxiliary decisions."

　　However, Wang Cong also pointed out that the data of intelligent algorithms comes from a variety of behaviors of consumers or users, and these behaviors themselves contain certain deviations. "If our algorithm feeds into this data, it may produce biased output. If consumers or users accept these algorithms, they may have a further biased impact on their decisions. The whole system is cyclical, so that the deviation is constantly amplified, so whether the economic benefits of the algorithm will be affected in this process will be the focus of our follow-up research. ”

　　If the training data of the recommendation system is known to have a certain deviation, how should the corresponding recommendation system be designed? "Eliminating bias and thus providing consumers with an unbiased recommendation tool is what we are currently doing research." Wang Cong introduced that regarding deviation correction, how to make end users truly accept the new algorithm, but also need to reduce people's resistance to the algorithm, which may require the mechanism design of the presentation of the algorithm results.

　　"It is very important to ensure the privacy of user data, I myself use the differential privacy framework, I need to do some parameter improvements in the training process of using the algorithm, by meeting the design of differential privacy, so as to improve the effect of the algorithm on privacy protection, and can ensure that the economic benefits of the algorithm are not too affected." Wang Cong said. (Zhongxin Jingwei APP)

　　Zhongxin Jingwei copyright, without written authorization, any unit and individual shall not reprint, excerpt or otherwise use.