Compressed Bayesian Federated Learning for Reliable Passive Radio Sensing in Industrial IoT

TL;DR

This paper tackles the high communication cost of decentralized Bayesian Federated Learning in industrial sensing by proposing compressed decentralized Bayesian FL (cd-bfl). The method injects compression via a quantization operator and allows multiple local updates before communication, drawing on Langevin-based sampling and consensus to approximate the global posterior $p(\boldsymbol{\theta}|\mathcal{D})$ without sharing raw data. On a case study of passive human localization with a network of IIoT radar devices, cd-bfl achieves substantial communication reductions (about $99\%$) while maintaining accuracy and, importantly, well-calibrated uncertainty estimates that remain robust under distribution shifts; it outperforms compressed frequentist FL in calibration. The work demonstrates the practical viability of reliable, communication-efficient decentralized Bayesian learning for safety-critical industrial applications and outlines future directions for theoretical analysis and optimization of the communication/computation balance.

Abstract

Bayesian Federated Learning (FL) has been recently introduced to provide well-calibrated Machine Learning (ML) models quantifying the uncertainty of their predictions. Despite their advantages compared to frequentist FL setups, Bayesian FL tools implemented over decentralized networks are subject to high communication costs due to the iterated exchange of local posterior distributions among cooperating devices. Therefore, this paper proposes a communication-efficient decentralized Bayesian FL policy to reduce the communication overhead without sacrificing final learning accuracy and calibration. The proposed method integrates compression policies and allows devices to perform multiple optimization steps before sending the local posterior distributions. We integrate the developed tool in an Industrial Internet of Things (IIoT) use case where collaborating nodes equipped with autonomous radar sensors are tasked to reliably localize human operators in a workplace shared with robots. Numerical results show that the developed approach obtains highly accurate yet well-calibrated ML models compatible with the ones provided by conventional (uncompressed) Bayesian FL tools while substantially decreasing the communication overhead (i.e., up to 99%). Furthermore, the proposed approach is advantageous when compared with state-of-the-art compressed frequentist FL setups in terms of calibration, especially when the statistical distribution of the testing dataset changes.

Figures (4)

• Figure 1: Bayesian fl setup: collaborating nodes ($N = 5$) iteratively exchange compressed samples via d2d communications and perform local computations to obtain a close approximation of the global posterior distribution.
• Figure 2: Human-Robot-Cooperative workspace scenario and FL setup.
• Figure 3: Comparison between dsgld and cd-bfl for varying values of $L$ in terms of validation accuracy (a) and ECE (b).
• Figure 4: Reliability plots attained by (a) dsgld, (b) cd-bfl and (c) cf-fl.