Parameter Oriented Scaling Consistency

Parameter-oriented scaling consistency (PSC) explores the inherent consistency in model predictions across different parameter scales, particularly useful for identifying anomalies like backdoor attacks in deep learning models. Current research focuses on leveraging PSC for improved model robustness and accuracy in various applications, including image segmentation, multi-view stereo reconstruction, and graph neural networks, often employing multi-scale consistency criteria and teacher-student training schemes. This research contributes to enhancing the reliability and performance of machine learning models, leading to more robust and accurate results in diverse fields such as medical image analysis and computer vision.