Researchers have developed a deepfake detection method focusing on a lightweight, low training complexity, and high-performance face biometrics technique with ideal size, weight, and power (SWaP).
Deepfake refers to artificial intelligence-synthesized, hyper-realistic video content that falsely depicts individuals saying or doing something. The researchers set out to tackle the significant threat that deepfake pose to our society and national security. The result is an innovative technological solution called DefakeHop for deepfake detection.
Combining machine learning, signal analysis, and computer vision techniques, the researchers developed an innovative theory and mathematical framework, Successive Subspace Learning, or SSL, as innovative neural network architecture. SSL is an entirely new mathematical framework for neural network architecture developed from signal transform theory. It offers a new signal representation and process that involves multiple transform matrices in a cascade. It is suitable for high-dimensional data with short-, mid-, and long-range covariance structures. It is a complete data-driven unsupervised framework that offers a brand new tool for image processing and understanding tasks such as face biometrics.
DefakeHop, according to the team, has several significant advantages over the current state-of-the-art, including being founded on a completely new SSL signal representation and transform theory. Because its internal modules and processing are explainable, it is mathematically transparent.
It is a weakly-supervised approach that provides a one-pass (no back-propagation) learning mechanism for labeling cost savings while requiring significantly less training complexity.
Related Content: Voice Biometrics To Prevent Frauds