Game-Theoretic Optimized Federated Learning for Heterogeneous IoT Object Detection

Area of Science:

• Computer Vision
• Artificial Intelligence
• Distributed Systems

Background:

• Federated object detection enables collaborative model training for IoT cameras without sharing raw data.
• Challenges include non-IID data, heterogeneous resources, unreliable links, and selfish client behavior.
• Existing methods struggle with efficiency and fairness in diverse IoT settings.

Purpose of the Study:

• To propose GO-FedDet, a game-theoretic framework for optimized federated object detection in heterogeneous IoT applications.
• To address limitations of non-IID scenes, resource inequality, intermittent connectivity, and selfish participation.
• To enhance accuracy, reduce communication overhead, and ensure long-term fairness in federated learning.

Main Methods:

• Formulating client participation as a Stackelberg game and client selection as an exact-potential resource game.
• Embedding game-theoretic equilibria into a proximal object detection objective.
• Implementing a utility-aligned aggregation mechanism to balance contribution, costs, and fairness.
• Proving the existence of a Stackelberg equilibrium and convergence properties of the selection game.

Main Results:

• GO-FedDet demonstrates improved detection accuracy compared to baseline federated methods.
• The framework significantly reduces communication costs in heterogeneous edge object detection scenarios.
• GO-FedDet achieves stabilized long-term fairness among participating clients.
• Theoretical analysis confirms convergence properties under challenging conditions like client drift and partial participation.

Conclusions:

• GO-FedDet provides an effective game-theoretic solution for optimizing federated object detection in heterogeneous IoT.
• The proposed framework successfully balances performance, efficiency, and fairness.
• This approach offers a robust method for collaborative learning in resource-constrained and dynamic edge environments.