Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Decentralized Fairness Aware Multi Task Federated Learning for VR Network

Krishnendu S. Tharakan, Carlo Fischione

TL;DR

This work tackles VR content delivery over wireless networks by proposing a decentralized, fairness-aware, multi-task federated learning framework (DMTFL) that personalizes per-base-station caching policies for tiled VR FOV content. By modeling unknown target distributions as mixtures and introducing cross-BS weighting, the approach achieves robustness against non-iid data while promoting fairness across BSs. The authors derive a PAC-type generalization bound based on Rademacher complexity and discrepancy, and provide a distributed optimization procedure to estimate discrepancies with minimal communication. Empirical results on realistic VR head-tracking data show substantial gains in cache hit rates and fairness compared to FedAvg, FedProx, and heuristic caching. The work offers a principled, scalable strategy for edge caching in MEC-enabled VR networks with theoretical guarantees and practical performance benefits.

Abstract

Wireless connectivity promises to unshackle virtual reality (VR) experiences, allowing users to engage from anywhere, anytime. However, delivering seamless, high-quality, real-time VR video wirelessly is challenging due to the stringent quality of experience requirements, low latency constraints, and limited VR device capabilities. This paper addresses these challenges by introducing a novel decentralized multi task fair federated learning (DMTFL) based caching that caches and prefetches each VR user's field of view (FOV) at base stations (BSs) based on the caching strategies tailored to each BS. In federated learning (FL) in its naive form, often biases toward certain users, and a single global model fails to capture the statistical heterogeneity across users and BSs. In contrast, the proposed DMTFL algorithm personalizes content delivery by learning individual caching models at each BS. These models are further optimized to perform well under any target distribution, while providing theoretical guarantees via Rademacher complexity and a probably approximately correct (PAC) bound on the loss. Using a realistic VR head-tracking dataset, our simulations demonstrate the superiority of our proposed DMTFL algorithm compared to baseline algorithms.

Decentralized Fairness Aware Multi Task Federated Learning for VR Network

TL;DR

This work tackles VR content delivery over wireless networks by proposing a decentralized, fairness-aware, multi-task federated learning framework (DMTFL) that personalizes per-base-station caching policies for tiled VR FOV content. By modeling unknown target distributions as mixtures and introducing cross-BS weighting, the approach achieves robustness against non-iid data while promoting fairness across BSs. The authors derive a PAC-type generalization bound based on Rademacher complexity and discrepancy, and provide a distributed optimization procedure to estimate discrepancies with minimal communication. Empirical results on realistic VR head-tracking data show substantial gains in cache hit rates and fairness compared to FedAvg, FedProx, and heuristic caching. The work offers a principled, scalable strategy for edge caching in MEC-enabled VR networks with theoretical guarantees and practical performance benefits.

Abstract

Wireless connectivity promises to unshackle virtual reality (VR) experiences, allowing users to engage from anywhere, anytime. However, delivering seamless, high-quality, real-time VR video wirelessly is challenging due to the stringent quality of experience requirements, low latency constraints, and limited VR device capabilities. This paper addresses these challenges by introducing a novel decentralized multi task fair federated learning (DMTFL) based caching that caches and prefetches each VR user's field of view (FOV) at base stations (BSs) based on the caching strategies tailored to each BS. In federated learning (FL) in its naive form, often biases toward certain users, and a single global model fails to capture the statistical heterogeneity across users and BSs. In contrast, the proposed DMTFL algorithm personalizes content delivery by learning individual caching models at each BS. These models are further optimized to perform well under any target distribution, while providing theoretical guarantees via Rademacher complexity and a probably approximately correct (PAC) bound on the loss. Using a realistic VR head-tracking dataset, our simulations demonstrate the superiority of our proposed DMTFL algorithm compared to baseline algorithms.

Paper Structure

This paper contains 7 sections, 1 theorem, 19 equations, 4 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Model and Problem Formulation
  3. DMTFL Algorithm
  4. Theoretical Guarantees & Proposed Algorithm
  5. Simulation Results
  6. Conclusion
  7. Proof of Theorem \ref{['thm:mainresult_cachingvr']}

Key Result

Theorem 1

(PAC Bound) Assuming the loss function is bounded, i.e., $\ell(a,b) < H, \forall a,b \in \mathcal{Y}$ then for any $\epsilon > 0$, with probability at least $1- \delta$, $\delta >0$, the following holds: where $\mathcal{P}(\mathbf{w}, \bm{\alpha}) = \sqrt{ \frac{1}{2} \sum_{b=1}^{B} \sum_{i=1}^{B} \left( \frac{w_b \alpha_{b,i}}{m_b} \right)^2 \log \left( \frac{|\Lambda_{\epsilon}|}{\delta} \right

Figures (4)

  • Figure 1: Tiled FOV representation based on a user’s FOV within the equirectangular (EQR) projection of a $360^{\circ}$ VR scene
  • Figure 2: Decentralized multi-task fair federated learning
  • Figure 3: Average cache hit versus cache size.
  • Figure 4: Minimum average cache hit versus cache size.

Theorems & Definitions (5)

  • Definition 1
  • Definition 2
  • Definition 3
  • Theorem 1
  • Remark 1