Linked Presentation: Gradients Look Alike: Sensitivity is Often Overestimated in DP-SGDMachine Learning needs Better Randomness Standards: Randomised Smoothing and PRNG-based attacks
