Rely on the robustness of trusted AI to effectively identify deep forgery and help banks fight identity fraud

Analysis of Yi Guan ：

With the development of computer vision and artificial intelligence , natural language processing 、 Image recognition 、 Face recognition and other technologies have been widely used in the financial field , Remote customer identification is used for credit card application 、 Mobile payment 、 Online loans and other businesses , It brings more convenience 、 More inclusive service experience . But at the same time , The forgery technology spawned by deep learning is also evolving and upgrading , Faced with such risk pressure ,AI The performance of robustness plays a key role , It's security AI An important ability to identify deep forgeries and have stable performance .

promote AI Robustness is an effective way to identify deep forgery

Deep forgery （Deepfake） It's deep learning （deep learning） And forgery （fake） A compound word of , It originally refers to the portrait synthesis technology based on deep learning , As technology evolves , Deep forgery has been developed to video 、 voice 、 The spoofing technology of multi-modal video forgery such as text and micro expression .

Due to the depth of the composite image 、 video 、 Audio 、 The production cost of text and other content is low 、 It's easy to operate 、 High fidelity , It is widely used in the identity fraud attack against banks in the network black industry ; Another aspect , Machine learning often requires a large number of high-quality training data to improve the accuracy of the model , however AI In the case of limited data , Its performance will be greatly limited , It is very difficult to distinguish the deeply forged content in various forms .

When the system is disturbed ,AI It should have good robustness （Robustness）, That is, the robustness of the system , Make it have the ability to resist external interference and attack , This is also AI The ability to survive in a risk environment , A credible AI Can be disturbed and uncertain , Ensure the stability of recognition rate and accuracy . Introduce artificial intelligence identification technology , utilize AI Robustness to resist deep forgery fraud , It can effectively improve the bank's ability to identify customers' identities and anti fraud ability .

Antagonism training can significantly improve the robustness of the model

Confrontation training is an effective method to resist confrontation attack , You can directly add confrontation samples to model training for learning , Get one “ Enhancement model ”, So as to defend against counter samples and forged samples , Improve AI For new data 、 Generalization ability of new scenarios . At present, the domestic artificial intelligence field has made some achievements in robustness research , And it can be applied to customer identification in the financial field 、 License identification 、 Anti fraud 、 Model evaluation and other scenarios .

Select the corresponding technical route according to the type of disturbance

Bank technology users are using AI Carry on voice 、 Text or image identification , It is necessary to distinguish whether the type of disturbance is benign or malicious , Thus, the corresponding technical route can be selected to improve AI Credibility of .

Goodwill disturbance is caused by poor quality of original data due to objective environmental reasons , Lead to AI The model decision-making of has produced deviation and error . In the actual application scenario of the bank , Intelligent customer service needs to face different customers 、 Photos of different lighting environments 、 Different language expression habits, etc , When the data to be processed changes slightly , Lacking robustness AI Recognition ability and accuracy will be significantly reduced . In this case , It is suggested to improve the quality of the image itself through technical means , The original data is transformed into a format that the model can understand before recognition ; in addition , Combine knowledge driven and data driven in model training , By extracting 、 After expression, it is calculated with a large amount of data , Form a more accurate model .

Malicious perturbations are deliberate forgery of images through deep synthesis , Or add specific noise to the real image to generate counter samples , A fraudulent attack on a bank . In the face of this type of disturbance , It is recommended that technical users choose AI Detecting forged samples against attack and defense tools , Identify whether the image has been tampered with , Establish appropriate risk control strategies for real-time early warning or interception ; When selecting counterfeit detection products , It is necessary to consider the rejection rate and accuracy rate of the product against different types of countermeasures , And data type compatibility 、 Performance in terms of operation speed, etc , Conduct a comprehensive assessment in combination with the risk control requirements of different business scenarios .

Notice of declaration ：

The third-party data and other information cited by Analysys in this article are from public sources , Analysys analysis assumes no responsibility for this . In any case , This article is for reference only , Not as any basis . The copyright of this article belongs to the publisher , Without authorization from Analysys , It is strictly prohibited to reprint 、 Reference or in any way use Analysys to analyze any content published . Any media authorized 、 When using the website or individual, the original text should be quoted and the source should be indicated , And the analysis point of view is based on the official content of Yiguan analysis , No form of deletion shall be made 、 Add 、 Splicing 、 deductive 、 Distortion, etc . Disputes over improper use , Yi Guan analysis does not assume any responsibility for this , And reserve the right to investigate the responsibility of the relevant subject .