Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Distributed Learning for Wi-Fi AP Load Prediction

Dariush Salami, Francesc Wilhelmi, Lorenzo Galati-Giordano, Mika Kasslin

TL;DR

This work tackles Wi-Fi AP load prediction in a setting with multiple independently managed deployments by applying Federated Learning (FL) and a data-free Knowledge Distillation KD-gen approach to a real campus dataset. The study evaluates not only predictive accuracy but also communication overhead, energy consumption, and training time, demonstrating that distributed learning can achieve up to $93\%$ accuracy gains and reduce costs by up to $80\%$ versus centralized training. KD-gen is shown to offer a compelling balance between accuracy and deployment cost, particularly for longer-horizon predictions, while FL provides strong performance for shorter horizons with lower communication. The contributions include a holistic assessment of deployment feasibility, including generator-based data synthesis for regression and practical considerations like energy and sustainability, underscoring the potential of privacy-preserving, collaborative Wi-Fi management across heterogeneous deployments.

Abstract

The increasing cloudification and softwarization of networks foster the interplay among multiple independently managed deployments. An appealing reason for such an interplay lies in distributed Machine Learning (ML), which allows the creation of robust ML models by leveraging collective intelligence and computational power. In this paper, we study the application of the two cornerstones of distributed learning, namely Federated Learning (FL) and Knowledge Distillation (KD), on the Wi-Fi Access Point (AP) load prediction use case. The analysis conducted in this paper is done on a dataset that contains real measurements from a large Wi-Fi campus network, which we use to train the ML model under study based on different strategies. Performance evaluation includes relevant aspects for the suitability of distributed learning operation in real use cases, including the predictive performance, the associated communication overheads, or the energy consumption. In particular, we prove that distributed learning can improve the predictive accuracy centralized ML solutions by up to 93% while reducing the communication overheads and the energy cost by 80%.

Distributed Learning for Wi-Fi AP Load Prediction

TL;DR

This work tackles Wi-Fi AP load prediction in a setting with multiple independently managed deployments by applying Federated Learning (FL) and a data-free Knowledge Distillation KD-gen approach to a real campus dataset. The study evaluates not only predictive accuracy but also communication overhead, energy consumption, and training time, demonstrating that distributed learning can achieve up to accuracy gains and reduce costs by up to versus centralized training. KD-gen is shown to offer a compelling balance between accuracy and deployment cost, particularly for longer-horizon predictions, while FL provides strong performance for shorter horizons with lower communication. The contributions include a holistic assessment of deployment feasibility, including generator-based data synthesis for regression and practical considerations like energy and sustainability, underscoring the potential of privacy-preserving, collaborative Wi-Fi management across heterogeneous deployments.

Abstract

The increasing cloudification and softwarization of networks foster the interplay among multiple independently managed deployments. An appealing reason for such an interplay lies in distributed Machine Learning (ML), which allows the creation of robust ML models by leveraging collective intelligence and computational power. In this paper, we study the application of the two cornerstones of distributed learning, namely Federated Learning (FL) and Knowledge Distillation (KD), on the Wi-Fi Access Point (AP) load prediction use case. The analysis conducted in this paper is done on a dataset that contains real measurements from a large Wi-Fi campus network, which we use to train the ML model under study based on different strategies. Performance evaluation includes relevant aspects for the suitability of distributed learning operation in real use cases, including the predictive performance, the associated communication overheads, or the energy consumption. In particular, we prove that distributed learning can improve the predictive accuracy centralized ML solutions by up to 93% while reducing the communication overheads and the energy cost by 80%.
Paper Structure (14 sections, 3 figures, 2 tables, 1 algorithm)

This paper contains 14 sections, 3 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. System Model
  4. ML-based Wi-Fi Load Prediction
  5. Distributed Training
  6. Federated learning
  7. Knowledge distillation with generative model
  8. Experimental Setup & Methodology
  9. Data
  10. Evaluation Scenario
  11. ML Model
  12. Performance: Accuracy, Overheads, Energy, and Time
  13. Results and Discussion
  14. Conclusions

Figures (3)

  • Figure 1: KD-gen detailed operation. The numbered bullets are aligned with the steps described in Algorithm \ref{['alg:distributed_training']}.
  • Figure 2: Histograms and curves for load of five different
  • Figure 3: of {ICL, DSCL, FL, KD-gen} at $s\in\{1,5,15,30\}$.