Adaptive Sampling for Private Worst-Case Group Optimization

TL;DR

This paper introduces ASC, a novel differentially private algorithm that adaptively samples and clips groups to improve worst-case group accuracy while maintaining consistent privacy guarantees.

Contribution

ASC is the first method to adaptively control sampling and clipping for worst-case group optimization under differential privacy, enhancing accuracy and privacy consistency.

Findings

Lower-variance gradients with ASC

Tighter privacy guarantees achieved

Significantly higher worst-case group accuracy

Abstract

Models trained by minimizing the average loss often fail to be accurate on small or hard-to-learn groups of the data. Various methods address this issue by optimizing a weighted objective that focuses on the worst-performing groups. However, this approach becomes problematic when learning with differential privacy, as unequal data weighting can result in inhomogeneous privacy guarantees, in particular weaker privacy for minority groups. In this work, we introduce a new algorithm for differentially private worst-case group optimization called ASC (Adaptively Sampled and Clipped Worst-case Group Optimization). It adaptively controls both the sampling rate and the clipping threshold of each group. Thereby, it allows for harder-to-learn groups to be sampled more often while ensuring consistent privacy guarantees across all groups. Comparing ASC to prior work, we show that it results in lower-variance gradients, tighter privacy guarantees, and substantially higher worst-case group accuracy without sacrificing overall average accuracy.