The deep integration of technology brings both light and shadows to human life, as it also fuels hidden growth. Technology risk has emerged as a major concern across various industries. From telecom fraud and fishing Trojans, to the buying and selling of personal information, and even today’s organized “wool parties,†cybercriminals continue to exploit vulnerabilities, which in turn demands advanced security and risk management solutions. The evolution of analytical technologies is essential to stay ahead of these threats.
In recent years, the rapid development of big data and artificial intelligence has become a powerful tool for risk control and anti-fraud efforts. DataVisor, a four-year-old company, has positioned itself with the slogan of "unsupervised learning algorithms." By combining supervised learning and an automatic rules engine, it offers protection against multiple scenarios, such as fake account registration, account theft, fraudulent transactions, identity theft, money laundering, counterfeit assessments, spam, and fake installation promotions.
Founded by Dr. Yinglian Xie, who holds a Ph.D. in computer science from Carnegie Mellon University, the company has over a decade of experience in cybersecurity, particularly in large-scale online attacks. Prior to founding DataVisor, she worked at Microsoft's Silicon Valley Research Lab. Recently, she shared her insights in an in-depth interview with the media.
**Three Major Technologies to Build an AI Moat**
According to Professor Deng Zhidong from Tsinghua University, the development of the AI industry involves four dimensions: scenarios, big data, computing power, and algorithms. While big data is foundational, and computing power is essential, the real challenge lies in the algorithm and talent. Once the specific scenarios are defined, big data becomes crucial, requiring the deep involvement of industry experts. Through data cleaning and labeling, their knowledge is transferred to machines, allowing AI to stand on the shoulders of giants.
However, in practice, many industries have accumulated vast amounts of data, but tagged data remains limited. This poses challenges due to manpower constraints and the rarity of labeled data. Moreover, there is often a lag in detecting new types of attacks, which directly impacts the model’s effectiveness. Traditional supervised learning struggles to keep up, prompting the need for unsupervised learning.
Unsupervised learning enables the automatic discovery of new attacks without relying on labels or training data. It can track evolving threats in real-time. As Xie Yinglian explained, "The biggest advantage is that it 'sees the enemy before they run,' either before or at the same time as the attack occurs," helping detect latent accounts and providing early warnings.
DataVisor analyzes three main types of data: account registration details, user behavior patterns, and other metadata like IP addresses, geographic locations, and device information. By clustering users who exhibit similar behaviors, the platform builds detailed user profiles. For instance, when a new user registers, the system may not find much information, but by analyzing all users, it can identify similarities in avatars, names, or phone models, highlighting suspicious activity.
Xie noted that while unsupervised machine learning is still underutilized, its practical application faces challenges in designing effective algorithms and architectures. Another key component is the automated rules engine, which uses machine learning to detect fraud groups and translate findings into understandable rules that meet regulatory requirements.
These three technologies—supervised learning, unsupervised learning, and the automated rules engine—work together, each playing a unique role. Supervised learning identifies regular patterns in labeled data, while the rules engine ensures transparency and reduces manual errors.
To support these technologies, DataVisor created the Global Smart Reputation Library, which extracts and integrates attack signals for deeper analysis. With data from over 2 billion users across different fields, including IP addresses, UA information, email domains, and device types, the database serves as a critical resource.
Based on this foundation, DataVisor developed a universal user analysis platform capable of adapting to various scenarios. This flexibility allows it to integrate with different data sources and applications, resulting in eight distinct use cases.
**Entering China, Investing in Finance**
In practical applications, DataVisor works closely with customers to tailor solutions. The first step is data preparation, ensuring quality input for accurate algorithm performance. Xie emphasized that while data comprehensiveness and accuracy remain challenges, organizations generally have some data infrastructure to work with.
Next, the team focuses on understanding customer needs and pain points, combining their algorithms with client data to solve real-world issues. After technical debugging and refinement based on feedback, the product is launched.
Customers can access test results through the DataVisor interface, user console, or via API for real-time or batch processing. They can also purchase custom rules to build their own systems. Deployment options include on-premise, SaaS, and private cloud, depending on business needs.
DataVisor’s clients include major platforms like Yelp, Pinterest, and Fortune 500 financial institutions. After entering the Chinese market in 2016, partnerships expanded to companies like Cheetah Mobile and Today’s Headlines.
Looking ahead, the company plans to focus more on the financial sector, offering services in account protection, credit applications, transaction settlements, and anti-money laundering. For example, one Fortune 500 company reduced fraud losses by over 30% using DataVisor’s early detection capabilities.
Another U.S. payment platform used DataVisor’s risk analysis tools to block 17% of transaction disputes, saving merchants over $50,000 annually.
Despite the crowded domestic market, Xie remains optimistic about healthy competition. She believes different players will play unique roles, with some focusing on whitelisting and Others on algorithmic solutions, all contributing to a stronger ecosystem.
With a growing team of Chinese engineers, DataVisor is positioning itself as a key player in the Chinese market. Future plans include enhancing unsupervised learning for broader applications, localizing features like Chinese language processing, and adapting to China-specific threats such as large-scale brushing and fake activities.
**Productivity**
Reflecting on her journey, Xie recalls her time at Microsoft, where she realized the potential of unsupervised learning. She noticed that many online behaviors, such as fraud and brushing, stemmed from account-level manipulation. This insight led her to develop a comprehensive approach to address fraud throughout the account lifecycle.
Her motivation came from a desire to go beyond traditional methods and create innovative solutions. Alongside her co-founder and CTO Yu Hao, who also came from Microsoft’s Silicon Valley lab, she aimed to build a robust system that could adapt to evolving threats.
Xie describes the anti-fraud industry as dynamic, with adversaries constantly changing. This makes the pursuit of new technologies both challenging and exciting. Over the past decade, this drive has fueled her commitment to the field, and she sees no end in sight.
Nd Auto Starter Engine Armature, Nd Starter Parts Armature, Nd Starter Parts Engine Armature,Nd Auto Starter Motor
YIWU JINGHONG AUTO PARTS CO.,LTD , https://en.jhauto.ru