Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Resource-Aware Hierarchical Federated Learning in Wireless Video Caching Networks

Md Ferdous Pervej, Andreas F. Molisch

TL;DR

A novel resource-aware hierarchical federated learning (RawHFL) solution for predicting user’s future content requests and extensive simulation results validate the proposed algorithm’s superiority, in terms of test accuracy and energy cost, over existing baselines.

Abstract

Backhaul traffic congestion caused by the video traffic of a few popular files can be alleviated by storing the to-be-requested content at various levels in wireless video caching networks. Typically, content service providers (CSPs) own the content, and the users request their preferred content from the CSPs using their (wireless) internet service providers (ISPs). As these parties do not reveal their private information and business secrets, traditional techniques may not be readily used to predict the dynamic changes in users' future demands. Motivated by this, we propose a novel resource-aware hierarchical federated learning (RawHFL) solution for predicting user's future content requests. A practical data acquisition technique is used that allows the user to update its local training dataset based on its requested content. Besides, since networking and other computational resources are limited, considering that only a subset of the users participate in the model training, we derive the convergence bound of the proposed algorithm. Based on this bound, we minimize a weighted utility function for jointly configuring the controllable parameters to train the RawHFL energy efficiently under practical resource constraints. Our extensive simulation results validate the proposed algorithm's superiority, in terms of test accuracy and energy cost, over existing baselines.

Resource-Aware Hierarchical Federated Learning in Wireless Video Caching Networks

TL;DR

A novel resource-aware hierarchical federated learning (RawHFL) solution for predicting user’s future content requests and extensive simulation results validate the proposed algorithm’s superiority, in terms of test accuracy and energy cost, over existing baselines.

Abstract

Backhaul traffic congestion caused by the video traffic of a few popular files can be alleviated by storing the to-be-requested content at various levels in wireless video caching networks. Typically, content service providers (CSPs) own the content, and the users request their preferred content from the CSPs using their (wireless) internet service providers (ISPs). As these parties do not reveal their private information and business secrets, traditional techniques may not be readily used to predict the dynamic changes in users' future demands. Motivated by this, we propose a novel resource-aware hierarchical federated learning (RawHFL) solution for predicting user's future content requests. A practical data acquisition technique is used that allows the user to update its local training dataset based on its requested content. Besides, since networking and other computational resources are limited, considering that only a subset of the users participate in the model training, we derive the convergence bound of the proposed algorithm. Based on this bound, we minimize a weighted utility function for jointly configuring the controllable parameters to train the RawHFL energy efficiently under practical resource constraints. Our extensive simulation results validate the proposed algorithm's superiority, in terms of test accuracy and energy cost, over existing baselines.
Paper Structure (26 sections, 5 theorems, 68 equations, 10 figures, 4 tables, 2 algorithms)

This paper contains 26 sections, 5 theorems, 68 equations, 10 figures, 4 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. State-of-the-Art
  3. Our Contributions
  4. System Model
  5. Network Model
  6. Content Request Model and Dataset Acquisition
  7. Content Request Model
  8. Dataset Acquisition Method
  9. Hierarchical Federated Learning: Preliminaries
  10. Resource-Aware Hierarchical Federated Learning: Convergence Analysis
  11. Resource-Aware Hierarchical Federated Learning Model
  12. Convergence of RawHFL
  13. RawHFL: Joint Problem Formulation and Solutions
  14. Joint Problem Formulation
  15. Problem Transformations and Solution
  16. ...and 11 more sections

Key Result

Theorem 1

Suppose $\eta < \mathrm{min}\left\{\frac{1}{2\sqrt{5} \beta \mathrm{L}}, \frac{1}{\beta E \mathrm{L}} \right\}$ and the above assumptions hold. Then, the average global gradient norm from $K$ global rounds of rawhfl is upper-bounded as where the expectations depend on clients' randomly selected mini-batches and $\mathrm{1}_{u,\mathrm{sc}}^{k,e}$'s. Besides, $\Omega^k \coloneqq \sum_{e=0}^{E-1} \s

Figures (10)

  • Figure 1: Network system model: ue are connected to their serving bs, and these bs are connected to their isp cloud network, while the csp has an agreement with the isp that allows placing one es to each bs
  • Figure 2: Privacy protection in different nodes: ($1$) ue uses tag ID, ($2$) ue sends encrypted content request to the serving bs, ($3$) bs sends the encrypted content request to its cloud for charging/authentication, ($4$) isp cloud sends encrypted content request information to csp, ($5$) csp decode actual content ID from the tagged ID, ($6$) csp sends encrypted information for the es to the isp cloud, ($7$) isp sends the encrypted information to serving bs, ($8$) serving bs forwards encrypted information to the es, ($9$) es decode the csp's information, ($10$) es sends encrypted video payload to serving bs, ($11$) bs sends the packet to the ue, and ($12$) ue decrypt the packet
  • Figure 3: Time flow in the proposed system model: each edge round can have multiple content request slots, while local training happens in between two consecutive edge rounds, and each global round consists of multiple edge rounds
  • Figure 4: Locations of the ue and bs in one realization
  • Figure 5: Clients' participation eligibility and number of selections in $400$ edge rounds
  • ...and 5 more figures

Theorems & Definitions (15)

  • Theorem 1
  • proof : Sketch of Proof
  • Remark 1
  • Corollary 1
  • Remark 2
  • Remark 3: Choice of the learning rate $\eta$
  • Remark 4
  • Remark 5
  • Remark 6
  • Theorem 1
  • ...and 5 more