What type of study is this?

This is a Experimental Study study.

October 3, 2025Open Access

On the Performance of Differentially Private Optimization with Heavy-Tail Class Imbalance

View Full Paper

Ask AI

Bookmark

Optimization with gradient descent suffers under heavy-tail class imbalance, impacting learning for low-frequency classes.
In contrast, DP-AdamBC significantly improves performance, with accuracy increases of approximately 8% for rare classes in controlled settings.
Using second-order optimization techniques helps mitigate the adverse effects of heavy-tail distributions on training accuracy.
Careful algorithm selection is crucial for better fitting and accuracy when engaging with imbalanced datasets.

No takes yet. Share an insight, caveat, or question.

Analysis reveals differential privacy learning struggles with low-frequency classes, suggesting improved algorithms may enhance performance.

Tang et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark

Analysis reveals differential privacy learning struggles with low-frequency classes, suggesting improved algorithms may enhance performance.