Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DNA: Differentially private Neural Augmentation for contact tracing

Rob Romijnders, Christos Louizos, Yuki M. Asano, Max Welling

TL;DR

The paper tackles privacy-preserving decentralized contact tracing by combining a traditional statistical SEIR-based inference with a neural augmentation component that is provably differentially private. It introduces Differentially Private Neural Augmentation (DNA), a Lipschitz-constrained DeepSet-based network that augments the FN-based inference, with formal DP guarantees under a three-level privacy hierarchy. In OpenABM-Covid19 simulations, DNA yields a lower peak infection rate at a rigorous DP budget of $\varepsilon=1$ per message and improves predictive accuracy (AUC $77.0 \to 83.1$) compared with DP-only baselines. The work demonstrates a practical pathway to harness neural augmentation for better pandemic mitigation while preserving essential privacy, and outlines avenues for addressing privacy composition and biases in iterative decentralized inference.

Abstract

The COVID19 pandemic had enormous economic and societal consequences. Contact tracing is an effective way to reduce infection rates by detecting potential virus carriers early. However, this was not generally adopted in the recent pandemic, and privacy concerns are cited as the most important reason. We substantially improve the privacy guarantees of the current state of the art in decentralized contact tracing. Whereas previous work was based on statistical inference only, we augment the inference with a learned neural network and ensure that this neural augmentation satisfies differential privacy. In a simulator for COVID19, even at epsilon=1 per message, this can significantly improve the detection of potentially infected individuals and, as a result of targeted testing, reduce infection rates. This work marks an important first step in integrating deep learning into contact tracing while maintaining essential privacy guarantees.

DNA: Differentially private Neural Augmentation for contact tracing

TL;DR

The paper tackles privacy-preserving decentralized contact tracing by combining a traditional statistical SEIR-based inference with a neural augmentation component that is provably differentially private. It introduces Differentially Private Neural Augmentation (DNA), a Lipschitz-constrained DeepSet-based network that augments the FN-based inference, with formal DP guarantees under a three-level privacy hierarchy. In OpenABM-Covid19 simulations, DNA yields a lower peak infection rate at a rigorous DP budget of per message and improves predictive accuracy (AUC ) compared with DP-only baselines. The work demonstrates a practical pathway to harness neural augmentation for better pandemic mitigation while preserving essential privacy, and outlines avenues for addressing privacy composition and biases in iterative decentralized inference.

Abstract

The COVID19 pandemic had enormous economic and societal consequences. Contact tracing is an effective way to reduce infection rates by detecting potential virus carriers early. However, this was not generally adopted in the recent pandemic, and privacy concerns are cited as the most important reason. We substantially improve the privacy guarantees of the current state of the art in decentralized contact tracing. Whereas previous work was based on statistical inference only, we augment the inference with a learned neural network and ensure that this neural augmentation satisfies differential privacy. In a simulator for COVID19, even at epsilon=1 per message, this can significantly improve the detection of potentially infected individuals and, as a result of targeted testing, reduce infection rates. This work marks an important first step in integrating deep learning into contact tracing while maintaining essential privacy guarantees.
Paper Structure (15 sections, 5 theorems, 21 equations, 3 figures, 1 table, 1 algorithm)

This paper contains 15 sections, 5 theorems, 21 equations, 3 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Method
  4. Statistical model
  5. Statistical messages
  6. Differential privacy
  7. Neural Augmentation
  8. Experimental details
  9. Experimental results
  10. Discussion and conclusion
  11. Appendix
  12. Notation
  13. Background
  14. Global sensitivity proof
  15. Experimental details

Key Result

Theorem 1

Following notation in Section sec:method and assuming that each message $\mu_{i}$ is bounded in the interval $[0, \gamma_u]$, for any two adjacent datasets as defined in Equation eqn:sensitivity_definition, the sensitivity for the FN inference function, $F(\cdot)$, is defined by:

Figures (3)

  • Figure 1: Three levels of DP analysis.
  • Figure 2: The trade-off between peak infection rate (y-axis) and the $\varepsilon$ DP parameter (x-axis). At the crucial setting of $\varepsilon=1$, our method, DNA, achieves a significantly lower peak infection rate.
  • Figure 3: This diagram illustrates the proof setup in Section \ref{['app:global_sens']}. The value of $\mu_1$ on day $\tau$ occurs in the conditional probability distribution $p(z_{\tau+1} | z_{\tau} = S, z_{N(u)})$. Filled lines indicate a transition to an equal state, and dashed lines indicate a transition to the next state in the $S \rightarrow E \rightarrow I \rightarrow R$ order.

Theorems & Definitions (5)

  • Theorem 1
  • Theorem 2
  • Lemma 1
  • Lemma 2
  • Lemma 3