Data Perturbation

Noise Injection

Process of controlled addition of random perturbations to data or training gradients to prevent inference attacks while maintaining model convergence.

← Indietro