Ali and Tsinghua promote industry-university-research cooperation to promote AI and security development

Elementary school students use photos to break Fengchao's face recognition system, self-driving cars run out of control, and AI's face change will make you a protagonist in a second ... With the rapid development, popularization and application of artificial intelligence technology, the subsequent security, Ethics and other issues have also caused widespread concern.

"Artificial intelligence security, as an important part of the research and industrialization of artificial intelligence, should promote deeper and wider cooperation between the government, academia and industry to systematically deploy the security of artificial intelligence," on December 8, in At the Symposium on AI and Security organized by Tsinghua University, Xue Qikun, vice president of Tsinghua University and academician of the Chinese Academy of Sciences, said: "This field is developing rapidly. We must grasp the right direction and key technologies at the critical moment. AI and security conferences can make new progress and gains. "

It is reported that the seminar was jointly organized by Tsinghua University, Ali Security, and Beijing Zhiyuan Artificial Intelligence Research Institute, and invited scientific researchers from universities, Internet companies, government departments and other parties to participate, aiming to increase the importance of AI and security from all walks of life. More security AI results.

"On the day of Tmall Double 11 in 2019, our security AI brain completed 38.8 billion risk prevention and control, and 99% of cattle were accurately intercepted by the system," said Qian Lei, general manager of security technology at Alibaba Group. For complex business scenarios, Tsinghua has strong technical and R & D capabilities. In the future, the two parties will strengthen cooperation in production, teaching, and research through talents and projects, which will be a very good exploration and proposition.

Liu Yiqun, Secretary of the Party Committee of the Computer Department of Tsinghua University, said that Ali and Tsinghua have been working closely together. This meeting will promote deeper cooperation between all parties, promote the establishment of a security management system in the field of artificial intelligence in China, and will also be more effective in the field of international artificial intelligence security To make a Chinese voice.

Consensus development of credible and interpretable security AI

The rapid and important development of artificial intelligence is self-evident. In 2019, the United States, Europe, Japan and other countries have introduced various policies and national strategies to promote the development of artificial intelligence; public data show that in 2019, the total number of artificial intelligence-related companies in China exceeded 2,500. By 2020, the scale of China's artificial intelligence industry Will be close to 50 billion yuan.

"In various scenarios, security and medical applications are more extensive. Recently, a camera on the Shanghai Metro caught a 20-year-old murderer, which is a typical application." Li Xinshe, deputy director of the National Industrial Information Security Development Center, pointed out in the report, but artificial intelligence is currently Facing data, algorithm, network, information, ethics and other security issues, "the establishment of credible artificial intelligence will become an important consideration for development."

As early as 1987, he led Tsinghua to establish the nation's first national key laboratory named after artificial intelligence to witness and participate in the development of the first and second generation of artificial intelligence technology in China. Zhang, Dean of the Institute of Artificial Intelligence at Tsinghua University, and Academician of the Chinese Academy of Sciences钹 In the keynote report, he emphasized that the development of a safe, reliable, and trusted third-generation artificial intelligence must be developed, so that the artificial intelligence technology can be widely and safely applied. The key to the development of technology lies in basic research. Key investments and breakthroughs must be made in basic research.

Ali and Tsinghua promote industry-university-research cooperation to promote AI and security development

(Zhang Huan, Dean of the Institute of Artificial Intelligence, Tsinghua University, Academician of the Chinese Academy of Sciences, delivered a keynote speech at the meeting)

"A photo of a bird with a little noise can make AI recognize a tiger," said Academician Zhang Yan. The security problem of artificial intelligence stems from the underlying defects of deep learning technology. It uses a very simple neural network system. Probabilistic methods to extract low-level features, and the correct method should be to extract high-level semantic features like humans. "We are working in this direction, but the reality is that there is no system that is always safe, and do n’t Expect an AI system that is always safe. Only by strengthening basic research can we have a forward-looking judgment and take timely measures. "

Xu Zongben, an academician of the Chinese Academy of Sciences and director of the National Engineering Laboratory for Big Data Algorithms and Analytical Technology, also emphasized in his keynote speech: the future development of artificial intelligence, especially applied technologies in key areas such as medical care, need to make full use of tools such as mathematics to strengthen basic research , Break through the bottleneck of accuracy, break data silos, and achieve small samples and interpretable deep learning.

AI and security development complement each other

Aiming at the many technical fields of AI and security, Zhang Weiming, deputy dean of the School of Cyberspace Security, University of Science and Technology of China, Wu Jun, deputy dean of the Institute of Cyberspace Security Technology, Shanghai Jiaotong University, and Shen Chao, deputy dean of cyberspace security at Xi'an Jiaotong University. Professor Huang Kaizhu of the Department of Electrical and Electronic Engineering of Xi'an Jiaotong-Liverpool University, and Zhang Chao, an associate professor of Tsinghua University, gave wonderful speeches at the meeting and provided suggestions from various angles, such as preventing the abuse and misuse of AI, strengthening the development of relevant standards, and strengthening Talent training, further deepening the integration of artificial intelligence and various physical industries, etc.

"When the industry is still worried about the security of AI, we have fully used AI technology in Ali's security scene. Under the attack and defense attributes of security itself, AI is also evolving," Qian Lei said. , Business environment governance, network security, data security, new retail security and other areas of AI applications, hoping to create a more secure, more trusted, interpretable, secure AI.

Ali and Tsinghua promote industry-university-research cooperation to promote AI and security development

(Qian Lei, General Manager of Security Technology of Alibaba Group, shared Alibaba's practice of using AI to solve security problems)

Qian Lei released a set of security AI control data at the scene: In terms of cracking down on fakes, 96% of suspected infringing links were blocked as soon as they went online, and 96% of intellectual property complaints were processed within 24 hours. On the other hand, it should respond to the identification and judgment of tens of millions of risk orders every day, and accumulate millions of malicious reviews.

Academician Xu Zongben commented that AI counterfeiting provides more machine learning and big data analysis, which is of great significance for authenticity detection. AI counterfeiting is a digital technology and artificial intelligence application scenario. "Ali has done a good job in this regard."

Frozen three feet is not a day cold. The Alibaba Group Security Department was established in 2009. The earliest stage is the expert operation stage, that is, through offline data analysis and judgment, to repair product vulnerabilities. "Until 2013, this was basically the human flesh model, but with the continuous expansion of Ali's business system If every business requires two or three experts to analyze, it is basically impossible, and we will enter the era of data intelligence later. "Qian Lei said that this period is characterized by the establishment of a comprehensive security system through a large amount of information, The indicator system, through people to interpret the changes in indicators, has become a rule landing, online prevention and control, "people are still the dominant."

"The essence of security risk control is a process of identification, determination, and disposal. We try to improve the efficiency through algorithms and use it in some traffic services. The effect is good. From 2018, we have fully shifted to the era of security AI," Qian Lei said. The level of security capability and how much data can be used in the scene are positively related. The large-scale application of AI technology in the security field will greatly improve the level of data used by security services. It also provides rich and changeable scenarios for the development of AI technology, and promotes the rapid development of AI technology. The two complement each other. "

In June 2019, Ali and Tsinghua University launched the Security AI Challenger Program, which conducted adversarial sample tests for face recognition, image classification, and other scenarios. A total of more than 2,000 teams signed up. The top five winning teams were also included in this meeting. A live award was given.

"We hope to cooperate with more parties to produce more results in the interpretability and security of AI and in more application scenarios to promote the development of artificial intelligence and the security industry." Qian Lei said.