The number of videos, pictures, and text data uploaded through the Internet every day can not be described as "massive"-more than 1.5 billion, and it is still increasing by orders of magnitude.
It filters out illegal content such as terror, violence, pornography, etc., so that it does not pollute or threaten cyberspace. In the past, artificial review was used. Now artificial intelligence (AI) has begun to take the stage.
On the 21st, the "2018 Cyber ​​Security Ecological Summit" jointly organized by the Internet Society of China and Alibaba, etc. opened in Beijing. AI's important role in facilitating network content security has become one of the hot topics of this summit.
It can be said that in the face of massive Internet data, using artificial intelligence to manage Internet data is a choice that has to be made.
"Compared with AI technology, the existing manual review methods are at a significant disadvantage in terms of cost and efficiency." The "AI Technology Empowers Network Content Security Assurance Research Report" released by the Internet Society of China at the summit (hereinafter referred to as the "Report" ") said.
According to the "Report", artificial intelligence has played an important role in network text content detection, text classification technology, video and picture content recognition, and voice content detection.
For example, the Cloud Shield content security system developed by Alibaba can provide intelligent identification services of multimedia content risks such as text, pictures, videos, etc. based on deep learning technology, effectively reducing information involving terror, violence, pornography and other illegal and illegal information.
According to Xu Jie, a senior engineer of the National Internet Emergency Response Center, traditional network security governance methods involve manual modeling and feature extraction, that is, describing something and forming a feature vector. The more complex things are, the more complex the feature vector increases exponentially.
"AI technology, especially the recently popular deep learning technology, provides new possibilities for us to deal with these problems." Xu Jie said, because deep learning technology no longer needs to manually describe the characteristics of things, but directly related things The pictures or videos of the images or videos are given to the machine to learn, and the machine builds the model by itself.
Xue Hui, a senior algorithm expert at Alibaba Group, believes that in addition to the above advantages, AI technology has also increased the reliability of network content governance. Because the experience of manual review varies from person to person, and this experience is more difficult to pass on, but AI technology is easier to inherit, whether it is a model or data.
Need to be more reliableAlthough AI has shown various advantages in ensuring the security of online content, as an emerging technological means, it is inevitable that there are still many shortcomings.
"AI itself has some problems. For example, AI is largely based on models and algorithms, and models and algorithms themselves are fragile." said Yang Zhengjun, secretary general of the Mobile Security Alliance, if its models and algorithms are attacked, they will fail. In addition, AI is trained based on massive amounts of data. If the data is contaminated or acquired by an attacker, the consequences will be unimaginable.
Li Jianfeng, a technical expert from iFlytek, believes that the biggest challenge of AI in network content governance is that the capabilities of AI need to be improved.
For example, Li Jianfeng said that some users sent Weibo to express their feelings, and the content was "the night club will pass, and the day will always come." This sentence could not be sent out repeatedly, and the user was puzzled. Later, a friend reminded him that it was because a "nightclub" appeared in it.
"Now we use AI technology, even if it is a very simple technology, it is difficult to be 100% correct." Li Jianfeng explained, because the current machine learning belongs to statistical machine learning, including deep learning, neural networks, etc. are based on Probability theory cannot guarantee 100% correctness in theory.
Door Bell Switch,Doorbell Switch,Doorbell Switch With Light,Doorbell Button Cover Plate
ZHEJIANG HUAYAN ELECTRIC CO.,LTD , https://www.huayanelectric.com