Inexact Unlearning Needs More Careful Evaluations to Avoid a False Sense of Privacy

no code implementations • 2 Mar 2024 • Jamie Hayes, Ilia Shumailov, Eleni Triantafillou, Amr Khalifa, Nicolas Papernot

In the privacy literature, this is known as membership inference.

Layer-Stack Temperature Scaling

no code implementations • 18 Nov 2022 • Amr Khalifa, Michael C. Mozer, Hanie Sedghi, Behnam Neyshabur, Ibrahim Alabdulmohsin

Inspired by this, we show that extending temperature scaling across all layers improves both calibration and accuracy.

Improving the Post-hoc Calibration of Modern Neural Networks with Probe Scaling

no code implementations • 29 Sep 2021 • Amr Khalifa, Ibrahim Alabdulmohsin

We present "probe scaling": a post-hoc recipe for calibrating the predictions of modern neural networks.