CG-FKAN: Compressed-Grid Federated Kolmogorov-Arnold Networks for Communication Constrained Environment

2025-11-03

AI Summary

Federated learning (FL) is widely deployed in privacy-sensitive applications, yet its inherent black-box nature limits model interpretability. While Kolmogorov–Arnold networks (KANs) enhance interpretability via learnable spline-based representations, their adaptive grid expansion incurs prohibitive communication overhead in bandwidth-constrained FL settings. To address this, we propose, for the first time, a communication-efficient federated KAN framework featuring grid compression and sparse coefficient selection—preserving spline interpretability while drastically reducing parameter transmission. We design the first FL training protocol explicitly incorporating communication budget constraints and theoretically derive an upper bound on the approximation error induced by grid compression. Extensive experiments demonstrate that, under identical communication budgets, our method reduces RMSE by 13.6% on average compared to fixed-grid KANs, achieving a superior trade-off among model accuracy, interpretability, and communication efficiency.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models
• Multiagent Systems: Agent Communication

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Federated learning (FL), widely used in privacy-critical applications, suffers from limited interpretability, whereas Kolmogorov-Arnold Networks (KAN) address this limitation via learnable spline functions. However, existing FL studies applying KAN overlook the communication overhead introduced by grid extension, which is essential for modeling complex functions. In this letter, we propose CG-FKAN, which compresses extended grids by sparsifying and transmitting only essential coefficients under a communication budget. Experiments show that CG-FKAN achieves up to 13.6% lower RMSE than fixed-grid KAN in communication-constrained settings. In addition, we derive a theoretical upper bound on its approximation error.

Problem

• Reduces communication overhead in federated KANs
• Compresses extended grids under communication constraints
• Maintains accuracy while minimizing transmitted coefficients

Innovation

• Compresses extended grids via sparsification
• Transmits only essential coefficients under budget
• Reduces communication overhead in federated learning