Label Proportion

Learning from Label Proportions (LLP) is a weakly supervised learning paradigm where models are trained using only the aggregate class proportions within groups of instances (bags), rather than individual instance labels. Current research focuses on improving the accuracy and efficiency of LLP methods, particularly for large bags and tabular data, employing techniques like contrastive learning, debiasing methods, and belief propagation. LLP's significance lies in its ability to address privacy concerns and data scarcity in various applications, including user modeling, medical image analysis, and remote sensing, by enabling model training with aggregated data.