Data sanitization with privacy guarantees

The researchers propose a data sanitization method with privacy guarantees that keeps classifiers trained on sanitized data close to classifiers trained on the original data, balancing privacy protection and data usability. Theoretical analysis shows that the risk of this method is lower than that of traditional input perturbation methods, and the advantage becomes more pronounced with more training samples. The method was validated on 12 benchmark datasets for its effectiveness in privacy-sensitive learning.