Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Personalized Over-the-Air Federated Learning with Reflecting Intelligent Surfaces

Jiayu Mao, Aylin Yener

TL;DR

This work tackles robust, scalable federated learning over wireless networks by leveraging RIS to enable over-the-air model aggregation. It develops ROAR-Fed for error-free downlink and APAF with personalization (PROAR-PFed) to adapt RIS phase shifts, local update counts, and transmission powers under time-varying imperfect CSI and non-i.i.d. data. A rigorous convergence analysis reveals a bound on the average gradient norm comprising seven error sources, and experiments on MNIST and Fashion-MNIST show RIS-enhanced OTA-FL consistently outperforms state-of-the-art baselines, with personalization gains when using personal RIS. The results underscore the practical impact of joint cross-layer optimization for RIS-assisted OTA-FL in 6G-like networks, enabling efficient global training and tailored per-user learning. Overall, the paper contributes theoretical foundations and algorithmic strategies for integrating RIS into personalized, over-the-air federated learning in heterogeneous edge environments.

Abstract

Over-the-air federated learning (OTA-FL) unifies communication and model aggregation by leveraging the inherent superposition property of the wireless medium. This strategy can enable scalable and bandwidth-efficient learning via simultaneous transmission of model updates using the same frequency resources, if care is exercised to design the physical layer jointly with learning. In this paper, a federated learning system facilitated by a heterogeneous edge-intelligent network is considered. The edge users (clients) have differing user resources and non-i.i.d. local dataset distributions. A general non-convex learning objective is considered for the model training task(s) at hand. We augment the network with Reconfigurable Intelligent Surfaces (RIS) in order to enhance the learning system. We propose a cross-layer algorithm that jointly assigns communication, computation and learning resources. In particular, we adaptively adjust the number of local steps in conjunction with RIS configuration to boost the learning performance. Our system model considers channel noise and channel estimation errors in both the uplink (model updates) and downlink (global model broadcast), employing dynamic power control for both. We provide the convergence analysis for the proposed algorithms and extend the frameworks to personalized learning. Our experimental results demonstrate that the proposed algorithms outperform the state-of-the-art joint communication and learning baselines.

Adaptive Personalized Over-the-Air Federated Learning with Reflecting Intelligent Surfaces

TL;DR

This work tackles robust, scalable federated learning over wireless networks by leveraging RIS to enable over-the-air model aggregation. It develops ROAR-Fed for error-free downlink and APAF with personalization (PROAR-PFed) to adapt RIS phase shifts, local update counts, and transmission powers under time-varying imperfect CSI and non-i.i.d. data. A rigorous convergence analysis reveals a bound on the average gradient norm comprising seven error sources, and experiments on MNIST and Fashion-MNIST show RIS-enhanced OTA-FL consistently outperforms state-of-the-art baselines, with personalization gains when using personal RIS. The results underscore the practical impact of joint cross-layer optimization for RIS-assisted OTA-FL in 6G-like networks, enabling efficient global training and tailored per-user learning. Overall, the paper contributes theoretical foundations and algorithmic strategies for integrating RIS into personalized, over-the-air federated learning in heterogeneous edge environments.

Abstract

Over-the-air federated learning (OTA-FL) unifies communication and model aggregation by leveraging the inherent superposition property of the wireless medium. This strategy can enable scalable and bandwidth-efficient learning via simultaneous transmission of model updates using the same frequency resources, if care is exercised to design the physical layer jointly with learning. In this paper, a federated learning system facilitated by a heterogeneous edge-intelligent network is considered. The edge users (clients) have differing user resources and non-i.i.d. local dataset distributions. A general non-convex learning objective is considered for the model training task(s) at hand. We augment the network with Reconfigurable Intelligent Surfaces (RIS) in order to enhance the learning system. We propose a cross-layer algorithm that jointly assigns communication, computation and learning resources. In particular, we adaptively adjust the number of local steps in conjunction with RIS configuration to boost the learning performance. Our system model considers channel noise and channel estimation errors in both the uplink (model updates) and downlink (global model broadcast), employing dynamic power control for both. We provide the convergence analysis for the proposed algorithms and extend the frameworks to personalized learning. Our experimental results demonstrate that the proposed algorithms outperform the state-of-the-art joint communication and learning baselines.

Paper Structure

This paper contains 21 sections, 4 theorems, 62 equations, 7 figures, 1 table, 3 algorithms.

Table of Contents

  1. Introduction
  2. System Model
  3. Federated Learning Model
  4. RIS-Assisted Communication Model
  5. Uplink Communication Model
  6. Downlink Communication Model
  7. Joint Communication and Learning Design
  8. Power Control for Uplink
  9. Phase Design
  10. Power Control for Downlink
  11. Computation Complexity
  12. Convergence Analysis
  13. Extension to Personalization
  14. Numerical Results
  15. Set Up
  16. ...and 6 more sections

Key Result

theorem 1

Let the learning rate be a constant, i.e., $\eta_t = \eta \leq \frac{1}{L}$. Set $P_t^i=P_i, \forall t \in [T]$. Under Assumptions a_smooth- a_bounded, we have: where $\frac{1}{\beta_i^2} = \frac{1}{T} \sum_{t=0}^{T-1} \frac{1}{(\beta_t^i)^2}$, $\frac{1}{\bar{\beta}^2} = \frac{1}{T} \sum_{t=0}^{T-1} \frac{1}{(\beta_t^u)^2}$, and

Figures (7)

  • Figure 1: The RIS-assisted communication system.
  • Figure 2: The personal RIS-assisted communication system.
  • Figure 3: Test accuracy on the MNIST dataset.
  • Figure 4: Test accuracy on Fashion-MNIST with $\gamma = 0.5$.
  • Figure 5: Test accuracy on Fashion-MNIST with $\gamma = 1$.
  • ...and 2 more figures

Theorems & Definitions (8)

  • theorem 1: Convergence Analysis
  • proof : Proof
  • corollary 1
  • proof : Proof
  • theorem 2
  • proof : Proof
  • corollary 2
  • proof : Proof