The deep application of technology is to bring "light" to human beings, and also to "dark" growth. Technology risk has become one of the major risks in various industries. From telecom fraud, to fishing Trojans, buying and selling personal information, to today's organized wool party, fraudsters have been exploiting attacks and areas, which also require security and risk. The change of analytical technology.
In recent years, the continuous development of big data and artificial intelligence technology has gradually become a powerful weapon for risk control and anti-fraud practitioners. The four-year-old DataVisor has launched the banner of “unsupervised learning algorithmâ€, combined with supervised learning and automatic rules engine to provide customers with protection for multiple application scenarios, including a large number of fake account registration, account theft, fraudulent transactions, identity theft, Money laundering transactions, counterfeit assessments, spam, fake installation promotion, etc.
Founder and CEO Yinglian Xie (Xie Yinglian) graduated from Carnegie Mellon University with a Ph.D. in computer science. She has more than 10 years of experience in the security field and has been working on large-scale online attacks. She previously worked for Microsoft Silicon Valley Research. hospital. Recently, she conducted an in-depth interview with the media.
Three major technologies to build a moat
"The development of artificial intelligence industry has four dimensions: scenarios, big data, computing power and algorithms. Big data is the foundation, computing power is the premise, and algorithms rely on talent. Big data is very important under the premise that the subdivision scenarios have been determined. This part requires the deep participation of top experts in the industry. Through the cleaning and labeling of big data, the knowledge of the top experts in the industry is transferred to the machine, so that artificial intelligence stands on the shoulders of giants.†Tsinghua professor Deng Zhidong told the media.
In reality, various industries may have more data accumulation, but the tagged data is very small, and it needs to rely on the deep participation of top experts in the industry. The manpower limitation is on the one hand, and on the other hand, the tagged data is rare. There is usually a hysteresis afterwards, and it is impossible to detect new types of unknown types of attacks. The timeliness and accuracy of the tag data directly affect the effect of the model. Unsupervised learning has been unable to meet the status quo, and practitioners have begun to apply unsupervised learning to deal with this situation.
As the name implies, unsupervised learning can automatically mine new attacks without relying on tags and training data. When the attack changes rapidly, it can automatically continue to track the mining. "The biggest advantage is that it 'passes the enemy to run', 'before the attack occurs or at the same time'," Xie Yinglian said, and can also detect the latency account, playing an early warning role.
According to reports, DataVisor generally mines three types of data for platform users: account registration information, behavior information, and other information (IP, geographic location, device, etc.). "Next, put the users who perform the same behavior on the platform for a period of time to detect, cluster analysis, and find similarities and correlations between accounts to form a single user portrait." For example, when a new user registers, The platform can't find more information, but when you contact all users, there may be some users who use very similar or similar avatars, names, phone models, etc., and the behavior is highlighted.
Xie Yinglian told reporters that the current unsupervised machine learning is less in practical applications. The difficulty lies in how to design algorithms, architecture and guarantee algorithms.
Another technology that is also under the unsupervised learning system is the automated rules engine. The traditional rule engine is manually debugged. Based on this, DataVisor uses machine learning technology to mine many fraud groups. Each group has one or more rules, so how to convert these results into human understanding. Rules to meet regulatory or other needs?
It is said that they will summarize the similarities of the rules and use the statistical principle to rigorously test the generation of rules, so that they are both explanatory and meet the needs of the platform.
“In general, these three technologies have different roles and complement each other. Supervised learning can unearth regular features in the case of labels, and can be combined with unsupervised learning. The automatic rules engine is mainly Meet interpretative requirements and reduce the cumbersome and error rate of manual debugging."
In addition, they created the DataVisor Global Smart Reputation Library to provide data support for these technologies. It mainly extracts and integrates attack signals and performs second-degree calculations to extract more representative signals. According to reports, the database has fraudulent data from more than 2 billion users in different fields, such as IP address, UA information, email domain name, device type and so on.
Based on the above three technologies plus the global smart reputation library, they developed a user analysis platform. Since the platform itself is universal and malleable, it can be hooked up with different data and different usage scenarios, and eight application scenarios emerge.
Entering China, investing in finance
So in practical applications, how does DataVisor combine application scenarios to provide services to customers?
Xie Yinglian said that the very important task in the first stage is to help customers sort out and clean the data. The quality of the data is also inseparable from the quality of the algorithm. Although DataVisor will face the challenge of data comprehensiveness and accuracy, she also pointed out that organizations have a strong sense of data, "there will be some (data combing) basis, although uneven."
The next step is to understand the customer's business scenarios and demand pain points. “Combine our algorithms and the other party's data to help customers solve practical problems.†After the above completion, the equivalent technical framework and products need to be debugged, based on customer feedback. Tune twice and then enter the product launch phase.
Customers can obtain test results through the DataVisor user interface, user analysis console, or batch export or real-time transfer of test results through the DataVisor results API, or directly purchase the rules to build their own books. The DataVisor data analytics platform offers a variety of deployment options, including on-premise, SaaS and private cloud deployments, depending on the customer's different business needs.
DataVisor's customers are said to be like the largest US review site Yelp, Pinterest, and Fortune 500 financial institutions. In November 2016, after officially entering the Chinese market, companies that established cooperation in China included public comment, cheetah mobile, and today's headlines.
DataVisor also said that the next step will be to force the financial industry. According to Xie Yinglian, the current cooperation with financial institutions mainly focuses on account protection, credit application, transaction settlement and anti-money laundering. Take the Fortune 500 companies as an example. The company has served in more than 200 countries and has been in the financial services industry for more than 100 years. DataVisor mainly provides anti-trafficking fraud services. The Director of Fraud and Risk Strategy can detect fraudsters a few days or hours before they launch an attack, reducing fraudulent transaction losses by more than 30%.
In addition, the largest merchant settlement payment platform in the United States uses DataVisor's one-stop risk data analysis platform to block 17% of transaction dispute fraud in real time, saving an average of more than 50,000 US dollars per year for platform merchants.
When it comes to the domestic wind control market, there is a lot of red seas, and the related technology providers are too numerous to mention. How does DataVisor across the ocean occupy a place in the competition?
"The market is vast, and there will inevitably be competition, but I think this will be a healthy competition." Xie Yinglian holds an optimistic view. "The different participants in the market will play different roles. Some of them are dedicated to whitelisting and fingerprint recognition. Types of work, some like we provide algorithms and platforms, are all improving the ecosystem."
She said that there are many Chinese engineers inside, and China is also the company's future strategic focus, and revealed the DataVisor domestic development plan. First, it will continue to improve the intelligence of unsupervised machine learning techniques, enabling them to match a wider range of scenarios and reduce manual intervention. Secondly, according to customer needs, localization adjustments, for example, optimization of Chinese language word processing; on the other hand, there are more wool party and brushing behaviors in China, and the scale is stronger, and China's attack characteristics will be adjusted.
Productivity
"After many years of experience in Microsoft, I realized the importance of unsupervised learning. Everyone thinks that the previous method is 'a headache, a painful foot,' and through the payment and brushing behavior on the Internet, we see that the essence is actually Fraud at the account level. So we came up with the idea of ​​solving the various frauds that existed in the account lifecycle."
Xie Yinglian described to reporters his mentality of starting a business. She said that Microsoft Research has a good research atmosphere, but for individuals, it is not satisfied with the local innovation through cooperation with various departments of Microsoft. Among them, "they" also includes co-founder and CTO Yu Hao, also from the Microsoft Silicon Valley Research Institute.
“One of the characteristics of the anti-fraud industry is that the opponents are constantly changing and the problem is not static. We are constantly pursuing new technologies to deal with attacks, and on the other hand, we are productizing this technological capability. Both processes are full of challenges. It also makes me very excited."
These may be summed up to support her drive for the industry over the past decade and for decades to come.
Insulated Power Cable,Bimetallic Crimp Lugs Cable,Pvc Copper Cable,Cable With Copper Tube Terminal
Taixing Longyi Terminals Co.,Ltd. , https://www.lycopperlugs.com