Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data Assetization via Resources-decoupled Federated Learning

Jianzhe Zhao, Feida Zhu, Lingyan He, Zixin Tang, Mingce Gao, Shiyu Yang, Guibing Guo

TL;DR

This work addresses data assetization under privacy constraints by proposing resource-decoupled federated learning (FL) involving a model owner, data owners, and computing centers. It introduces a Tripartite Stackelberg model solved by backward induction to reach a Stackelberg-Nash equilibrium ($SNE$) that maximizes global utility, and a dynamic algorithm, QD-RDFL, that evaluates data quality via training loss and adaptively adjusts incentives and matches using Gale-Shapley. Theoretical analysis guarantees $SNE$ existence and uniqueness, and the dynamic mechanism improves data-quality assessment and model performance. Empirical results on MNIST, CIFAR-10, and CIFAR-100 show enhanced global utility and incentivized collaboration, demonstrating practical potential for data assetization in privacy-preserving ML ecosystems.

Abstract

With the development of the digital economy, data is increasingly recognized as an essential resource for both work and life. However, due to privacy concerns, data owners tend to maximize the value of data through the circulation of information rather than direct data transfer. Federated learning (FL) provides an effective approach to collaborative training models while preserving privacy. However, as model parameters and training data grow, there are not only real differences in data resources between different data owners, but also mismatches between data and computing resources. These challenges lead to inadequate collaboration among data owners, compute centers, and model owners, reducing the global utility of the three parties and the effectiveness of data assetization. In this work, we first propose a framework for resource-decoupled FL involving three parties. Then, we design a Tripartite Stackelberg Model and theoretically analyze the Stackelberg-Nash equilibrium (SNE) for participants to optimize global utility. Next, we propose the Quality-aware Dynamic Resources-decoupled FL algorithm (QD-RDFL), in which we derive and solve the optimal strategies of all parties to achieve SNE using backward induction. We also design a dynamic optimization mechanism to improve the optimal strategy profile by evaluating the contribution of data quality from data owners to the global model during real training. Finally, our extensive experiments demonstrate that our method effectively encourages the linkage of the three parties involved, maximizing the global utility and value of data assets.

Data Assetization via Resources-decoupled Federated Learning

TL;DR

This work addresses data assetization under privacy constraints by proposing resource-decoupled federated learning (FL) involving a model owner, data owners, and computing centers. It introduces a Tripartite Stackelberg model solved by backward induction to reach a Stackelberg-Nash equilibrium () that maximizes global utility, and a dynamic algorithm, QD-RDFL, that evaluates data quality via training loss and adaptively adjusts incentives and matches using Gale-Shapley. Theoretical analysis guarantees existence and uniqueness, and the dynamic mechanism improves data-quality assessment and model performance. Empirical results on MNIST, CIFAR-10, and CIFAR-100 show enhanced global utility and incentivized collaboration, demonstrating practical potential for data assetization in privacy-preserving ML ecosystems.

Abstract

With the development of the digital economy, data is increasingly recognized as an essential resource for both work and life. However, due to privacy concerns, data owners tend to maximize the value of data through the circulation of information rather than direct data transfer. Federated learning (FL) provides an effective approach to collaborative training models while preserving privacy. However, as model parameters and training data grow, there are not only real differences in data resources between different data owners, but also mismatches between data and computing resources. These challenges lead to inadequate collaboration among data owners, compute centers, and model owners, reducing the global utility of the three parties and the effectiveness of data assetization. In this work, we first propose a framework for resource-decoupled FL involving three parties. Then, we design a Tripartite Stackelberg Model and theoretically analyze the Stackelberg-Nash equilibrium (SNE) for participants to optimize global utility. Next, we propose the Quality-aware Dynamic Resources-decoupled FL algorithm (QD-RDFL), in which we derive and solve the optimal strategies of all parties to achieve SNE using backward induction. We also design a dynamic optimization mechanism to improve the optimal strategy profile by evaluating the contribution of data quality from data owners to the global model during real training. Finally, our extensive experiments demonstrate that our method effectively encourages the linkage of the three parties involved, maximizing the global utility and value of data assets.
Paper Structure (18 sections, 4 theorems, 15 equations, 8 figures, 6 tables, 3 algorithms)

This paper contains 18 sections, 4 theorems, 15 equations, 8 figures, 6 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. Problem Statement
  5. Tripartite Stackelberg Model
  6. Analysis of Tripartite Stackelberg Model
  7. QD-RDFL
  8. Experiments
  9. Experiments Setting
  10. Experimental Results
  11. Conclusions and Future works
  12. Appendix
  13. Proof of Lemma 2
  14. Proof of Theorem 1
  15. Gale-Shapley Algorithm
  16. ...and 3 more sections

Key Result

Lemma 1

For computing center $C_m$, there is a unique optimal $d_m^*$ to maximize its utility $U_m$ for the given data quantity from the data owners.

Figures (8)

  • Figure 1: Data Assetaization via Resource-Decoupled FL
  • Figure 2: Uniqueness and existence of SNE of our method, (a) the utility of model owner with different $\eta$ and (b) the utility of data owner with different $q_n$.
  • Figure 3: Effect of initial $f_n$ to $U_s$ and the accuracy of global model on MNIST, (a) comparison of $U_s$ with different proportions of data owners and different report deviation ratio $f_n$, and (b) the accuracy of global model with different $f_n$.
  • Figure 4: $U_s$ with different number of data owners on different datasets.
  • Figure 5: Average of $U_n$ with different number of data owners on different datasets.
  • ...and 3 more figures

Theorems & Definitions (6)

  • Definition 1: Tripartite Stackelberg Model
  • Definition 2: Stackelberg-Nash equilibrium
  • Lemma 1
  • Lemma 2
  • Lemma 3
  • Theorem 1