[Paper Review] Game of Privacy: Towards Better Federated Platform Collaboration under Privacy Restriction

Chuhan Wu, Fangzhao Wu+5 more|arXiv (Cornell University)|Feb 10, 2022

Privacy-Preserving Technologies in Data4 citations

TL;DR

This paper proposes FedGame, a game-theoretic framework for reciprocal collaboration in vertical federated learning under privacy constraints. Platforms strategically allocate limited data quotas to maximize their model performance by estimating inter-platform data value via 'deposit' data and negotiating via gradient-based policy optimization, achieving effective multi-platform collaboration without centralized trust.

ABSTRACT

Vertical federated learning (VFL) aims to train models from cross-silo data with different feature spaces stored on different platforms. Existing VFL methods usually assume all data on each platform can be used for model training. However, due to the intrinsic privacy risks of federated learning, the total amount of involved data may be constrained. In addition, existing VFL studies usually assume only one platform has task labels and can benefit from the collaboration, making it difficult to attract other platforms to join in the collaborative learning. In this paper, we study the platform collaboration problem in VFL under privacy constraint. We propose to incent different platforms through a reciprocal collaboration, where all platforms can exploit multi-platform information in the VFL framework to benefit their own tasks. With limited privacy budgets, each platform needs to wisely allocate its data quotas for collaboration with other platforms. Thereby, they naturally form a multi-party game. There are two core problems in this game, i.e., how to appraise other platforms' data value to compute game rewards and how to optimize policies to solve the game. To evaluate the contributions of other platforms' data, each platform offers a small amount of "deposit" data to participate in the VFL. We propose a performance estimation method to predict the expected model performance when involving different amount combinations of inter-platform data. To solve the game, we propose a platform negotiation method that simulates the bargaining among platforms and locally optimizes their policies via gradient descent. Extensive experiments on two real-world datasets show that our approach can effectively facilitate the collaborative exploitation of multi-platform data in VFL under privacy restrictions.

Motivation & Objective

• To address the lack of incentives in vertical federated learning (VFL) where only the label-holding platform benefits.
• To model platform collaboration under privacy budget constraints as a multi-party game with reciprocal data sharing.
• To enable platforms to strategically allocate limited data quotas to maximize individual model performance gains.
• To design a trustless, decentralized mechanism that avoids reliance on a central authority for profit allocation.
• To develop a scalable method for estimating cross-platform data value and optimizing collaboration policies under real-world privacy restrictions.

Proposed method

• Each platform offers a small 'deposit' of its data to participate in inter-platform VFL training, enabling performance estimation without full data sharing.
• A multi-variable regression model estimates expected model performance gains based on varying amounts of inter-platform deposit data.
• Platforms simulate real-world bargaining through a negotiation process that optimizes collaboration policies via gradient descent.
• The game-solving framework uses performance estimation to compute rewards and guide policy updates toward Nash equilibrium.
• Collaboration policies are locally optimized based on estimated gains, enabling decentralized, incentive-aligned decision-making.
• The final model is trained after policies converge, ensuring effective use of privacy-constrained data across platforms.

Experimental results

Research questions

• RQ1 How can platforms be incentivized to contribute data in VFL when only the label-holding platform traditionally benefits?
• RQ2 How can inter-platform data value be estimated accurately under privacy constraints without full data sharing?
• RQ3 What negotiation mechanism enables stable, effective collaboration among multiple platforms with limited data budgets?
• RQ4 How does the choice of deposit data size affect the accuracy of performance estimation and final model performance?
• RQ5 What is the optimal balance between data contribution and privacy budget usage in reciprocal VFL collaboration?

Key findings

• FedGame enables all platforms to benefit from collaboration, avoiding the prisoner’s dilemma where no platform contributes data.
• On the Adult and CTR datasets, platforms strategically allocate data to those with the most valuable complementary features, such as the first platform prioritizing collaboration with the third platform.
• The method achieves optimal average model performance when the negotiation parameter γ is set to 2.5, balancing cooperation and competition.
• Using 5% of total data as deposit data yields the best trade-off between estimation accuracy and privacy budget consumption.
• Performance prediction errors are high with very small deposit sets due to insufficient data, and performance degrades with large deposits due to excessive privacy budget usage.
• The platform negotiation mechanism successfully avoids greedy behavior and maintains balanced, mutually beneficial collaborations across all platforms.