Skip to main content
. Author manuscript; available in PMC: 2024 Apr 1.
Published in final edited form as: Transact Mach Learn Res. 2023 Jun;2023:https://openreview.net/forum?id=K0CAGgjYS1.

Table 2:

Calibration metrics ECE and MCE by non-DP (no clipping) and DP optimizers.

ECE % MCE %
non-DP DP (small R) DP (large R) non-DP DP (small R) DP (large R)
CIFAR10 1.3 0.9 1.1 54.8 58.6 27.5
MNIST 0.4 2.3 0.7 49.3 56.2 33.4
SNLI 13.0 22.0 17.6* 34.7 62.5 28.9*
*

Note that the SNLI experiment uses the mix-up training as described in Section 5.5