Quadratic Privacy-Signaling Games and the MMSE Information Bottleneck Problem for Gaussian Sources

Ertan Kazıklı, Sinan Gezici, Serdar Yüksel

Introduction

In modern data-driven ecosystems, users frequently share information to access services, creating a fundamental tension between data utility and individual privacy. For instance, smart meter data provides utility for grid management but may inadvertently reveal private household routines. This paper investigates this conflict through a game-theoretic lens, specifically focusing on privacy-signaling games involving jointly Gaussian sources and quadratic cost functions.

Problem Formulation

The authors model a scenario where a sender observes two correlated Gaussian random vectors: $X$ (representing private information) and $Y$ (representing useful information). The sender's goal is to transmit a signal $Z$ that allows a receiver to accurately estimate $Y$ while minimizing the receiver's ability to estimate $X$. Conversely, the receiver aims to minimize the Mean Squared Error (MSE) for both $X$ and $Y$.

System models for (a) the general privacy-signaling game and (b) the Information Bottleneck variant.

Two primary game-theoretic frameworks are analyzed:

  1. Nash Equilibrium: A simultaneous-move game where neither player commits to a strategy in advance.
  2. Stackelberg Equilibrium (Bayesian Persuasion): A leader-follower game where the sender commits to an encoding strategy, and the receiver optimizes their decoding strategy in response.

Additionally, the paper addresses the MMSE Gaussian Information Bottleneck (IB) Problem, a variant where the sender only observes one of the variables (e.g., $X$), which is correlated with the target $Y$, as shown in the bottom diagram of the figure above.

Methodology

To tackle the multi-dimensional complexity, the researchers employ a linear transformation to decorrelate the Gaussian sources into independent scalar components. This simplification allows for a component-wise analysis of the equilibrium strategies.

A critical contribution of this work is the proof of linear optimality. While the players are allowed to choose any measurable (potentially non-linear) encoding and decoding functions, the authors demonstrate that for Gaussian sources under quadratic costs, the optimal strategies are inherently linear. This result is significant because signaling games often result in complex, non-linear "quantized" equilibria.

Key Insights and Results

Equivalence of Equilibria

The study reveals that the payoff-dominant Nash equilibrium—the equilibrium providing the best outcome for both parties—coincides with the Stackelberg equilibrium. This suggests that the ability to commit to a strategy does not necessarily grant the sender a better outcome than the natural equilibrium reached through mutual best-responses.

The "Silence" Threshold

The research identifies explicit conditions under which communication becomes non-informative. If the sender's privacy requirement ($\delta$) exceeds a specific threshold relative to the correlation between $X$ and $Y$, the optimal strategy for the sender is to transmit only noise or remain silent.

MMSE Information Bottleneck

The paper provides an explicit linear solution for the MMSE Gaussian IB problem. As privacy constraints tighten, the sender optimally "shuts down" data dimensions, prioritizing the suppression of components most strongly correlated with the private variable. These results extend to scenarios involving noisy Gaussian channels, where the sender simply scales their linear strategy to meet power constraints.

Limitations and Future Work

The findings are primarily constrained by the assumption of jointly Gaussian sources and quadratic (MSE) costs. While MSE is a standard metric, it may not fully capture all nuances of privacy, such as those addressed by Differential Privacy. Future research could explore multi-user extensions or non-Gaussian distributions where non-linear signaling might emerge as the superior strategy.

Conclusion

This work provides a rigorous mathematical foundation for understanding privacy-utility trade-offs in Gaussian environments. By proving the optimality of linear strategies and characterizing the resulting equilibria, it offers practical insights for designing privacy-preserving communication systems.

Summary updated

Abstract

We investigate a privacy-signaling game problem in which a sender with privacy concerns observes a pair of correlated random vectors which are modeled as jointly Gaussian. The sender aims to hide one of these random vectors and convey the other one whereas the objective of the receiver is to accurately estimate both of the random vectors. We analyze these conflicting objectives in a game theoretic framework with quadratic costs where depending on the commitment conditions (of the sender), we consider Nash or Stackelberg (Bayesian persuasion) equilibria. We show that a payoff dominant Nash equilibrium among all admissible policies is attained by a set of explicitly characterized linear policies. We also show that a payoff dominant Nash equilibrium coincides with a Stackelberg equilibrium. We formulate the information bottleneck problem within our Stackelberg framework under the mean squared error distortion criterion where the information bottleneck setup has a further restriction that only one of the random variables is observed at the sender. We show that this MMSE Gaussian Information Bottleneck Problem admits a linear solution which is explicitly characterized in the paper. We provide explicit conditions on when the optimal solutions, or equilibrium solutions in the Nash setup, are informative or noninformative.
BibTeX
@article{kazkl2022quadratic,
  author = {Ertan Kazıklı and Sinan Gezici and Serdar Yüksel},
  title = {{Quadratic Privacy-Signaling Games and the MMSE Information Bottleneck Problem for Gaussian Sources}},
  journal = {IEEE Transactions on Information Theory},
  year = {2022},
  doi = {10.1109/tit.2022.3177258},
}
Figure