Evidential Neural Network

Evidential neural networks (ENNs) are a class of deep learning models designed to quantify uncertainty in predictions, going beyond simple confidence scores. Current research focuses on improving ENN performance and reliability across various tasks, including multi-view learning, image retrieval, and time-series analysis, often employing architectures like transformers and leveraging belief theory frameworks such as subjective logic and Dempster-Shafer theory. This enhanced uncertainty quantification is crucial for building trustworthy AI systems, particularly in safety-critical applications like medical image analysis and autonomous navigation, where understanding and managing uncertainty is paramount. The integration of physics-based priors into ENNs is also an emerging area, aiming to improve robustness and generalization.