Trade-off Functions for DP-SGD with Subsampling based on Random Shuffling: Tight Upper and Lower Bounds
arXiv:2605.06259v2 Announce Type: replace Abstract: We derive a tight analysis of the trade-off function for Differentially Private Stochastic Gradient Descent (DP-SGD) with subsampling based on random shuffling within the $f$-DP framework. Our analysis covers the regime $\sigma \geq \sqrt{3/\ln M}$, where $\sigma$ is the noise multiplier and $M$ is the number of rounds within a single epoch. Unlike $f$-DP analyses for Poisson subsampling, which yield non-closed implicit formulas that can be machine computed but are non-transparent, random shuffling admits a tight analysis yielding transparent
This research provides a more transparent and tight analysis for a critical technique in privacy-preserving AI, emerging at a time when data privacy regulations and concerns are accelerating globally.
Improved understanding and implementation of differential privacy in AI directly impacts the trustworthiness, deployability, and regulatory compliance of AI systems, especially in sensitive domains.
The availability of tight analytical bounds for DP-SGD with random shuffling simplifies evaluation and potentially improves the practical deployment of differentially private machine learning algorithms.
- · AI researchers
- · Privacy-focused AI developers
- · Healthcare sector
- · Financial services
- · Organizations with weak privacy practices
More efficient and reliable differentially private AI models become feasible for real-world applications.
Increased adoption of privacy-preserving AI could lead to new regulatory standards and consumer expectations for data security.
The enhanced trust in AI systems due to provable privacy could accelerate AI integration into highly regulated and sensitive industries.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG