Negative Preference

Negative preference, encompassing the modeling and utilization of disliked or unwanted items/data, is a growing area of research across machine learning applications. Current efforts focus on incorporating negative information into model training, often using contrastive learning or specialized architectures like signed graph convolutional networks, to improve performance metrics such as accuracy and out-of-distribution generalization. This research is significant because effectively leveraging negative preferences enhances the accuracy and robustness of various systems, from medical diagnosis and recommendation systems to image-text matching and copy detection.