Bag Level

Bag-level analysis focuses on classifying groups of data points (bags) rather than individual instances, addressing challenges in scenarios with limited or aggregated labels. Current research emphasizes developing effective bag-level classifiers, often employing Multiple Instance Learning (MIL) techniques and incorporating architectures like transformers and graph neural networks to capture inter-instance relationships within bags. This approach is particularly relevant for applications like whole slide image classification in pathology and privacy-preserving machine learning where only aggregated data is available, offering improvements in efficiency and robustness.