Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Model Wireless Federated Learning with Downlink Beamforming

Chong Zhang, Min Dong, Ben Liang, Ali Afana, Yahia Ahmed

TL;DR

This work tackles wireless federated learning for concurrently training multiple ML models under noisy downlink/uplink channels. It introduces round-robin device-model scheduling and a downlink multi-group multicast beamforming design, deriving an upper bound on the per-model optimality gap ${\mathbb{E}}[\|\boldsymbol{\theta}_{m,SM}-\boldsymbol{\theta}_m^{\star}\|^2]$ that guides the beamforming design. The optimization reduces to minimizing the sum of inverse SINRs via a multi-group multicast formulation, solved efficiently with projected gradient descent. Simulations on MNIST show that the proposed MultiModel approach significantly outperforms sequential single-model training and multi-model ZF, achieving near $97\%$ accuracy in common wireless settings. This demonstrates a practical path toward scalable, faster convergence for multi-model FL in wireless networks.

Abstract

This paper studies the design of wireless federated learning (FL) for simultaneously training multiple machine learning models. We consider round robin device-model assignment and downlink beamforming for concurrent multiple model updates. After formulating the joint downlink-uplink transmission process, we derive the per-model global update expression over communication rounds, capturing the effect of beamforming and noisy reception. To maximize the multi-model training convergence rate, we derive an upper bound on the optimality gap of the global model update and use it to formulate a multi-group multicast beamforming problem. We show that this problem can be converted to minimizing the sum of inverse received signal-to-interference-plus-noise ratios, which can be solved efficiently by projected gradient descent. Simulation shows that our proposed multi-model FL solution outperforms other alternatives, including conventional single-model sequential training and multi-model zero-forcing beamforming.

Multi-Model Wireless Federated Learning with Downlink Beamforming

TL;DR

This work tackles wireless federated learning for concurrently training multiple ML models under noisy downlink/uplink channels. It introduces round-robin device-model scheduling and a downlink multi-group multicast beamforming design, deriving an upper bound on the per-model optimality gap that guides the beamforming design. The optimization reduces to minimizing the sum of inverse SINRs via a multi-group multicast formulation, solved efficiently with projected gradient descent. Simulations on MNIST show that the proposed MultiModel approach significantly outperforms sequential single-model training and multi-model ZF, achieving near accuracy in common wireless settings. This demonstrates a practical path toward scalable, faster convergence for multi-model FL in wireless networks.

Abstract

This paper studies the design of wireless federated learning (FL) for simultaneously training multiple machine learning models. We consider round robin device-model assignment and downlink beamforming for concurrent multiple model updates. After formulating the joint downlink-uplink transmission process, we derive the per-model global update expression over communication rounds, capturing the effect of beamforming and noisy reception. To maximize the multi-model training convergence rate, we derive an upper bound on the optimality gap of the global model update and use it to formulate a multi-group multicast beamforming problem. We show that this problem can be converted to minimizing the sum of inverse received signal-to-interference-plus-noise ratios, which can be solved efficiently by projected gradient descent. Simulation shows that our proposed multi-model FL solution outperforms other alternatives, including conventional single-model sequential training and multi-model zero-forcing beamforming.
Paper Structure (13 sections, 1 theorem, 15 equations, 2 figures)

This paper contains 13 sections, 1 theorem, 15 equations, 2 figures.

Table of Contents

  1. Introduction
  2. System Model
  3. Multi-Model FL System
  4. Wireless Communication Model
  5. Multi-Model Downlink-Uplink Transmissions
  6. Downlink Broadcast
  7. Local Model Update
  8. Uplink Aggregation
  9. Multi-Model Downlink Beamforming Design
  10. Convergence Rate Analysis
  11. Downlink Multi-Group Multicast Beamforming Design
  12. Simulation Results
  13. Conclusion

Key Result

Proposition 1

For the multi-model FL system described in Section sec:FL_alg, under Assumptions assump_smooth--assump_bound_diff and for $\eta_n<\frac{1}{\lambda}$, $\forall n$, the expected model optimality gap after $S$ frames is upper bounded by where $\Gamma_m \triangleq {\mathbb{E}}[\| \boldsymbol \theta_{m,0} - \boldsymbol \theta_m^\star\|^2]$, $G_n \triangleq 4(1-\eta_n\lambda)^{2J}$, $\Lambda \triangleq

Figures (2)

  • Figure 1: An example of round robin scheduling of device-model assignment in a frame for training 3 models.
  • Figure 2: Left: Test accuracy vs. $M$ (from Model A). Middle & Right: Test accuracy vs. communication round $t$: Middle -- Model A; Right -- Model B.

Theorems & Definitions (1)

  • Proposition 1