Alexander Benvenuti

Achieving an impressive 88.17% accuracy on CIFAR10 under differential privacy, this framework redefines the balance between data privacy and model performance.

Existing differential privacy methods struggle with symbolic trajectory data, but this new mechanism slashes error by up to 55% on real-world data.