Abstract:A series of recent works by Lyu, Wang, Vadhan, and Zhang (TCC `21, NeurIPS `22, STOC `23) showed that composition theorems for non-interactive differentially private mechanisms extend to the concurrent composition of interactive differentially private mechanism, when differential privacy is measured using $f$-DP and the adversary is adaptive. We extend their work to the $\textit{continual observation setting,}$ where the data is arriving online in a potentially adaptive manner. More specifically, we show that all composition theorems for non-interactive differentially private mechanisms extend to the concurrent composition of continual differentially private mechanism, where the adversary is adaptive. We show this result for $f$-DP, which also implies the result for pure DP and $(\epsilon, \delta)$-DP.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: How to extend the composition theorems of non - interactive differential privacy mechanisms to concurrent composition continuous differential privacy mechanisms in the continuous - observation setting, especially when facing adaptive adversaries. Specifically, the authors hope to prove that all composition theorems applicable to non - interactive differential privacy mechanisms can also be applied to concurrent - composition continuous differential privacy mechanisms, where data arrives online and may be updated in an adaptive manner. ### Problem Background Differential Privacy (DP) is a popular method for measuring the degree of privacy protection provided by an algorithm when conducting statistical analysis on sensitive data sets. For static data sets (i.e., batch - processing settings), a great deal of research has explored differential privacy mechanisms. However, when data sets change dynamically, especially when queries and data updates occur interleaved, this setting has only received more attention until recently. In 2010, Dwork et al. introduced the concept of "differential privacy under continuous observation" to describe mechanisms for protecting privacy in such a dynamic environment. ### Research Objectives The objective of this paper is to extend the existing composition theorems of non - interactive differential privacy mechanisms to concurrent - composition continuous differential privacy mechanisms. Specifically, the authors hope to prove the following: 1. **Concurrent - composition continuous differential privacy mechanisms**: All composition theorems applicable to non - interactive differential privacy mechanisms can be extended to concurrent - composition continuous differential privacy mechanisms, even in the face of adaptive adversaries. 2. **Scope of application**: These results are applicable not only to pure differential privacy (pure DP) and (ϵ, δ)-differential privacy, but also to the broader f - differential privacy (f - DP). ### Main Contributions - **Theoretical extension**: By extending the work of Lyu, Wang, Vadhan and Zhang et al., this paper proves that in the continuous - observation setting, all composition theorems of non - interactive differential privacy mechanisms are also applicable to concurrent - composition continuous differential privacy mechanisms. - **Privacy definitions**: The results of this paper cover multiple privacy definitions, including f - differential privacy, (ϵ, δ)-differential privacy and pure differential privacy. - **Technical tools**: This paper uses technical tools such as the Interactive Pre - Post - Processing Mechanism (IPPM) to ensure privacy protection in concurrent composition. Through these contributions, this paper provides an important theoretical basis and technical support for designing and analyzing differential privacy mechanisms in a dynamic data environment.

Concurrent Composition for Continual Mechanisms

Concurrent Composition of Differential Privacy

Fully Adaptive Composition in Differential Privacy

Composition in Differential Privacy for General Granularity Notions (Long Version)

Adaptive Privacy Composition for Accuracy-first Mechanisms

Privacy accounting $\varepsilon$conomics: Improving differential privacy composition via a posteriori bounds

Cactus Mechanisms: Optimal Differential Privacy Mechanisms in the Large-Composition Regime

Unified Enhancement of Privacy Bounds for Mixture Mechanisms via $f$-Differential Privacy

Constant matters: Fine-grained Complexity of Differentially Private Continual Observation

Avoiding Pitfalls for Privacy Accounting of Subsampled Mechanisms under Composition

Towards Separating Computational and Statistical Differential Privacy

[Effects of head down bed rest or drinking water containing electrolytic silver-ion on intestinal bifidobacteria].

Bounded and Unbiased Composite Differential Privacy

ConTPL: Controlling Temporal Privacy Leakage in Differentially Private Continuous Data Release

Convergent Differential Privacy Analysis for General Federated Learning: the $f$-DP Perspective

Almost Tight Error Bounds on Differentially Private Continual Counting

Breaking the Communication-Privacy-Accuracy Tradeoff with F-Differential Privacy

Elephants Do Not Forget: Differential Privacy with State Continuity for Privacy Budget

Two Views of Constrained Differential Privacy: Belief Revision and Update

Composition Properties of Inferential Privacy for Time-Series Data

A unified concurrent-composition method to state/event inference and concealment in discrete-event systems