Privileged Information

Learning using privileged information (LUPI) leverages auxiliary data available during training but not at inference time to improve model performance. Current research focuses on applying LUPI to diverse problems, including bias mitigation, domain generalization, and robust prediction in the presence of noisy labels, often employing teacher-student architectures or generative models to synthesize privileged information. This approach holds significant promise for enhancing model accuracy and robustness across various fields, particularly where data scarcity or quality issues are prevalent, such as medical imaging and autonomous driving.