Deepfakes, disinformation, AI, truth decay | Homeland Security Newswire

Breakthrough Technology a Game Changer for Deepfake Detection

Most current state-of-the-art techniques for deepfake video detection and media forensics methods are based on the deep learning mechanism, which have many inherent weaknesses in terms of robustness, scalability and portability, You said.

According to the team, DefakeHop has several significant advantages over current start-of-the-arts, including:

· It is built upon the entirely new SSL signal representation and transform theory. It is mathematically transparent since its internal modules and processing are explainable

· It is a weakly-supervised approach, providing a one-pass (without needing back-propagation) learning mechanism for the labeling cost saving with significantly lower training complexity

· It generates significantly smaller model sizes and parameters. Its complexity is much lower than that of state-of-the-art and it can be effectively implemented on the tactical edge devices and platforms

· It is robust to adversarial attacks. The deep learning based approach is vulnerable to adversarial attacks. This research provides a robust spatial-spectral representation to purify the adversarial inputs, thus adversarial perturbations can be effectively and efficiently defended against

This research supports the Army’s and lab’s AI and ML research efforts by introducing and studying an innovative machine learning theory and its computational algorithms applied to intelligent perception, representation and processing, You said.

“We expect future soldiers to carry intelligent yet extremely low size–weight–power vision-based devices on the battlefield,” You said. “Today’s machine learning solution is too sensitive to a specific data environment. When data are acquired in a different setting, the network needs to be re-trained, which is difficult to conduct in an embedded system. The developed solution has quite a few desired characteristics, including a small model size, requiring limited training data, with low training complexity and capable of processing low-resolution input images. This can lead to game-changing solutions with far reaching applications to the future Army.”

The researchers successfully applied the SSL principle to resolve several face biometrics and general scene understanding problems. Coupled with the DefakeHop work, they developed a novel approach called FaceHop based on the SSL principle to a challenging problem–recognition and classification of face gender under low image quality and low-resolution environments.

The team continues to develop novel solutions and scientific breakthroughs for face biometrics and for general scene understanding, for example, target detection, recognition and semantic scene understanding.

“We all have seen AI’s substantial impact on society–both good and bad, and AI is transforming many things,” Hu said. “Deepfake is an adverse example. The creation of sophisticated computer-generated imagery has been demonstrated for decades through the use of various visual effects in the entertainment industry, but recent advances in AI and machine learning have led to a dramatic increase in the realism of fake content and the ease of access to these tools.”

The research team has the opportunity to address these challenging issues, which have both military and every day impact.

“We see this research as new, novel, timely and technically feasible today,” You said. “It is a high risk, high innovation effort with transformative potential. We anticipate that this research will provide solutions with significant advantages over current techniques, and add important new knowledge to the sciences of artificial intelligence, computer vision, intelligent scene understanding and face biometrics.”

Breakthrough Technology a Game Changer for Deepfake Detection

Recent stories

Free Subscription