Shulou(Shulou.com)05/31 Report--

This article mainly explains "what is DefakeHop technology". The content of the explanation is simple and clear, and it is easy to learn and understand. Please follow the editor's train of thought to study and learn "what is DefakeHop technology".

Depth detection: using DefakeHop to detect Deepfake

ARL researchers Dr Suya You and Dr Shuowen (Sean) Hu point out that deep forgery is surreal video content synthesized by artificial intelligence, falsifying someone's words and deeds. Most of the latest deep forgery video detection and media forensics methods are based on deep learning, which has many inherent weaknesses in robustness, scalability and portability.

"due to the development of generative neural networks, AI-driven deep forgery has developed so rapidly that there is a lack of reliable technology for detection and defense," Dr. You said. "We urgently need to find new ways to understand the mechanisms behind the amazing performance of deep forgery, and to develop effective defense solutions with solid theoretical support."

By combining the experience of team members with machine learning, signal analysis and computer vision, researchers developed an innovative theoretical and mathematical framework-continuous subspace learning (SSL), which is an innovative neural network architecture. The researchers point out that SSL is a key innovation of DefakeHop.

Kuo pointed out: "SSL is a new mathematical framework for neural network architecture developed from signal conversion theory." It is completely different from the traditional method, it provides a new signal representation and process, which involves a number of cascaded transformation matrices. It is very suitable for high-dimensional data with short, medium and long-range covariance structures. SSL incorporates this feature in its design. It is a complete data-driven unsupervised framework that provides new tools for image processing and understanding tasks such as facial biometric recognition. "

You points out that most of the latest technologies for deep forgery video detection and media forensics are based on deep learning mechanisms.

DefakeHop is substantially ahead of the advanced level of the industry.

The researchers believe that the performance of DefakeHop is significantly ahead of the most advanced technology in the industry, mainly in the following four aspects:

DefakeHop is based on the new theory of SSL signal representation and transformation. It is mathematically transparent, and its internal modules and processing can be explained.

DefakeHop is a weakly supervised method, which provides a learning mechanism through (no back propagation), thus saving tagging costs and greatly reducing the complexity of training.

The size and parameters of the resulting model are much smaller. The complexity is much lower than the latest technology, and can be effectively implemented on tactical edge devices and platforms.

It has a powerful function against aggression. Methods based on deep learning are vulnerable to adversarial attacks. This study provides a robust spatial spectral representation to purify adversarial inputs, so it can effectively defend against adversarial disturbances.

Professor You pointed out:

This study provides support for AI and ML research work in the United States Army and laboratories by introducing and studying an innovative machine learning theory and its computational algorithms applied to intelligent perception, representation, and processing.

We hope that future soldiers will be able to carry lightweight intelligent vision devices on the battlefield. Today's machine learning solutions are too sensitive to specific data environments. When obtaining data with different settings, the network needs to be retrained, which is very difficult in embedded systems.

The solution currently developed has many ideal features, including small model size, low training data, low training complexity, and the ability to deal with low-resolution input images. This is a solution that can lead to a change in the rules of the game and is expected to be widely used in the army in the future.

Solved several problems of facial biometric recognition.

Researchers have successfully applied the SSL principle to solve a variety of facial biometric recognition and general scene understanding problems. Combined with the work of DefakeHop, they developed a novel method, FaceFop based on SSL principle, to solve the problem of problem recognition and face classification in challenging environments with low image quality and low resolution.

It is reported that the team will continue to develop innovative solutions for facial biometric recognition and general scene understanding, such as target detection, recognition and semantic scene understanding.

Professor Hu said: "We have all seen the significant impact of artificial intelligence on society, both good and dark, but there is no denying that artificial intelligence is changing a lot of things." Deep forgery is a negative example. Over the past few decades, the entertainment industry has shown us that computers can generate fake, highly complex visual effects, and now the latest technologies in artificial intelligence and machine learning have greatly reduced the technical barriers to content forgery. Related tools are also extremely easy to obtain. "

Finally, the research team believes that deep forgery technology has an impact on military and daily life (threat), while in the research field of countering deep forgery, DefakeHop scheme has significant advantages over existing technologies, which provides a new paradigm and knowledge for artificial intelligence research in artificial intelligence, computer vision, intelligent scene understanding and facial biometric recognition.

Thank you for your reading, the above is the content of "what is DefakeHop technology", after the study of this article, I believe you have a deeper understanding of what is DefakeHop technology, and the specific use needs to be verified in practice. Here is, the editor will push for you more related knowledge points of the article, welcome to follow!

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.