Fetching the paper…

Equivariant Differentially Private Deep Learning: Why DP-SGD Needs Sparser Models · Around