Ant Digital won the "Financial Technology Technology Innovation and Application Case" award for its Deepfake detection solution at the 12th Digital Finance and Technology Finance Conference of the Zhongguancun Forum Series. This solution relies on Ant Digital Tianji Laboratory to build the industry's first large-scale, high-quality, multi-modal Deepfake data set, effectively improving the accuracy and reliability of the Deepfake detection model, and successfully applied to actual financial services scenarios to effectively protect the security of user assets. The construction of this data set solves the previous problems of small training scale of Deepfake detection models in the financial field and inability to be evaluated in real environments, and promotes the development of traditional detection models.
Recently, at the 12th Digital Finance and Technology Finance Conference of the Zhongguancun Forum series of events, Ant Digital's Deepfake detection solution was selected into the conference's "Financial Technology Technology Innovation and Application Cases".
Relying on its subsidiary Tianji Lab, Ant Digital has built a large-scale, high-quality, multi-modal Deepfake data set for the first time in the industry. It synthesizes more than one million levels of multimedia content and fully simulates Deepfake in the real-world financial risk control environment. Attack samples have become an important criterion for evaluating the performance of existing deepfake detection models in the financial field. In financial business scenarios, Ant Digits’ Deepfake detection accuracy on multiple test data sets reached over 98%, and it successfully prevented many frauds using Deepfake technology and protected users’ asset security.
This data set solves the problem that deepfake detection models in the financial field cannot be trained on a large scale and cannot be evaluated in real environments. It also promotes the development of traditional detection models from the perspective of multi-modal analysis. At present, this data set has become the key capability of Ant Digital’s anti-deep fake product ZOLOZDeeper to serve external customers.
It is understood that Ant Digital uses up to 81 advanced deepfake technologies to generate high-quality synthetic images, covering a variety of forgery technology types, complex lighting conditions, background environments and facial expressions to simulate complex and realistic real-world attack environments. In addition to static images, a large amount of video data containing sound is also collected and generated, including more than 100 types of forgery techniques, covering different languages, accents and background noise, ensuring the diversity and complexity of the data set.
In the data preprocessing and annotation stage, Ant Digital cleans and preprocesses the collected data to ensure data quality. The expert team annotates the data to make it clear whether each image or video is generated by Deepfake, while ensuring that traces of forgery are minimized to achieve a highly realistic effect. Previously, Ant Digital released an AI data synthesis and production platform that achieved "AI dominance" at the data annotation level, reducing the amount of manual annotation the annotation model relied on by more than 70%.
In addition, Ant Digital launched a Deepfake Offensive and Defense Challenge at the 2024 Bund Conference, using the Deepfake data set as the basic training and testing data for the competition. It attracted more than 2,200 players from 26 countries and regions around the world to sign up for the competition. Through the algorithm solutions contributed by contestants, the attack quality and detection difficulty of the Deepfake data set have been effectively verified and evaluated.
With the development of artificial intelligence technology, Deepfake technology is also advancing rapidly. This technology uses deep learning algorithms to realistically replace faces in videos. Although Deepfake has active applications in fields such as entertainment and media, Deepfake technology brings new risks in the financial field, especially in identity verification and transaction verification. Financial institutions' identity verification systems often rely on biometric technologies such as facial recognition. Once these systems are deceived by Deepfake technology, serious financial fraud may result.
In view of this, it is very necessary to develop a detection system for Deepfake attacks in the financial field, but a powerful Deepfake detection and defense model requires a high-quality face Deepfake data set that conforms to the real-world environment, so how to build a data set that simulates the real world and how Verifying its effectiveness is an urgent issue.
Ant Digital's deepfake detection solution provides a strong guarantee for financial security. The large-scale high-quality data sets it constructs also provide valuable resources for industry research, promote the development of deepfake detection technology, and provide a new direction for future financial security. .