Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Toward a Sustainable Federated Learning Ecosystem: A Practical Least Core Mechanism for Payoff Allocation

Zhengwei Ni, Zhidu Li, Wei Chen, Zhaoyang Zhang, Zehua Wang, F. Richard Yu, Victor C. M. Leung

TL;DR

This work tackles the problem of fair and stable payoff allocation in federated learning, where the core of the coalition may be empty. It introduces a practical Least Core (LC) mechanism that minimizes the maximum dissatisfaction $e$ through an LP formulation under Efficiency and Stability, and couples it with a stack-based pruning scheme (SPriLC) to scale LC to large federations. Through intrusion-detection and vertical FL case studies, LC identifies indispensable participants and deters free-riders, outperforming data-volume and leave-one-out baselines and revealing distinct behavior from Shapley value in emphasizing indispensability. The approach includes guidelines for pruning thresholds and demonstrates scalability to networks with tens to hundreds of participants, enabling sustainable, stable cross-silo FL ecosystems in industrial settings.

Abstract

Emerging network paradigms and applications increasingly rely on federated learning (FL) to enable collaborative intelligence while preserving privacy. However, the sustainability of such collaborative environments hinges on a fair and stable payoff allocation mechanism. Focusing on coalition stability, this paper introduces a payoff allocation framework based on the least core (LC) concept. Unlike traditional methods, the LC prioritizes the cohesion of the federation by minimizing the maximum dissatisfaction among all potential subgroups, ensuring that no participant has an incentive to break away. To adapt this game-theoretic concept to practical, large-scale networks, we propose a streamlined implementation with a stack-based pruning algorithm, effectively balancing computational efficiency with allocation precision. Case studies in federated intrusion detection demonstrate that our mechanism correctly identifies pivotal contributors and strategic alliances. The results confirm that the practical LC framework promotes stable collaboration and fosters a sustainable FL ecosystem.

Toward a Sustainable Federated Learning Ecosystem: A Practical Least Core Mechanism for Payoff Allocation

TL;DR

This work tackles the problem of fair and stable payoff allocation in federated learning, where the core of the coalition may be empty. It introduces a practical Least Core (LC) mechanism that minimizes the maximum dissatisfaction through an LP formulation under Efficiency and Stability, and couples it with a stack-based pruning scheme (SPriLC) to scale LC to large federations. Through intrusion-detection and vertical FL case studies, LC identifies indispensable participants and deters free-riders, outperforming data-volume and leave-one-out baselines and revealing distinct behavior from Shapley value in emphasizing indispensability. The approach includes guidelines for pruning thresholds and demonstrates scalability to networks with tens to hundreds of participants, enabling sustainable, stable cross-silo FL ecosystems in industrial settings.

Abstract

Emerging network paradigms and applications increasingly rely on federated learning (FL) to enable collaborative intelligence while preserving privacy. However, the sustainability of such collaborative environments hinges on a fair and stable payoff allocation mechanism. Focusing on coalition stability, this paper introduces a payoff allocation framework based on the least core (LC) concept. Unlike traditional methods, the LC prioritizes the cohesion of the federation by minimizing the maximum dissatisfaction among all potential subgroups, ensuring that no participant has an incentive to break away. To adapt this game-theoretic concept to practical, large-scale networks, we propose a streamlined implementation with a stack-based pruning algorithm, effectively balancing computational efficiency with allocation precision. Case studies in federated intrusion detection demonstrate that our mechanism correctly identifies pivotal contributors and strategic alliances. The results confirm that the practical LC framework promotes stable collaboration and fosters a sustainable FL ecosystem.
Paper Structure (6 sections, 3 figures, 2 tables)

This paper contains 6 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Least Core in Federated Learning Network
  3. Implementation of Practical LC
  4. Commercial Case Study: Federated Intrusion Detection Model
  5. Least Core vs. Shapley Value
  6. Conclusions

Figures (3)

  • Figure 1: Implementation of the practical LC scheme. (Top) Steps 1-3, which use the LIFO principle to find tall informative coalitions. (Bottom) Step 4, illustrating the structure of our LP solver for payoff allocation.
  • Figure 2: Overview of the federated intrusion detection case study and payoff allocation results. (Top-left) The four participants with their distinct organizational profiles and data distributions. (Top-right) The F1-weighted performance scores for all 15 possible coalitions, which define the value of each partnership. (Bottom) A comparative analysis of final payoff allocations under three methods: allocation by data volume, the leave-one-out, and our proposed least core method.
  • Figure 3: The contribution of each participant evaluated by the LC and the SV.