Collaborative Metric Learning

Collaborative Metric Learning (CML) aims to improve recommendation systems and object retrieval by learning distance metrics that leverage collaborative information from user interactions or visual similarities. Recent research focuses on addressing limitations of existing CML methods, such as bias from imbalanced data or the computational cost of negative sampling, through techniques like incorporating multiple user representations to capture diverse preferences and developing sampling-free alternatives. These advancements enhance the accuracy and efficiency of CML, leading to improved performance in recommendation systems and general object retrieval tasks.