Architecture-Adaptive Uncertainty Fusion for Deepfake Detection
Researchers have developed a new framework called Correlation-Optimized Fusion (COF) to improve the reliability of uncertainty quantification in deepfake detection systems. COF adaptively fuses five different sources of uncertainty, requiring minimal computational resources for optimization. While existing methods perform better on in-domain benchmarks, COF demonstrates superior performance under distribution shifts, making it a practical tool for real-world forensic applications. AI
IMPACT Improves reliability of AI-driven deepfake detection in real-world scenarios by addressing distribution shift.