Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Distributed Hierarchical Machine Learning for Joint Resource Allocation and Slice Selection in In-Network Edge Systems

Sulaiman Muhammad Rashid, Ibrahim Aliyu, Jaehyung Park, Jinsul Kim

TL;DR

The paper tackles the challenge of joint wireless and computing resource management with optimal slice selection in a slice-enabled COIN-MEC edge network for latency-sensitive Metaverse applications. It formulates a mixed-integer nonlinear program (MINLP) and decomposes it into three subproblems (SP1-SP3), then trains a distributed hierarchical DeepSets-S model with a shared encoder and task-specific decoders to approximate solver-derived policies. The model enforces permutation equivariance over variable-size WD sets via a slack-aware normalization, achieving high accuracies ($Acc_1 = 95.26\%$, $Acc_2 = 95.67\%$, $Acc = 0.7486$, $Acc_{\text{bin}} = 0.8824$) and reducing online inference time by $86.1\%$, while staying within $6.1\%$ of the optimal cost. The work demonstrates scalable AI-native orchestration for 6G edge systems and outlines future directions for dynamic environments, semi-supervised learning, and real-world validation.

Abstract

The Metaverse promises immersive, real-time experiences; however, meeting its stringent latency and resource demands remains a major challenge. Conventional optimization techniques struggle to respond effectively under dynamic edge conditions and high user loads. In this study, we explore a slice-enabled in-network edge architecture that combines computing-in-the-network (COIN) with multi-access edge computing (MEC). In addition, we formulate the joint problem of wireless and computing resource management with optimal slice selection as a mixed-integer nonlinear program (MINLP). Because solving this model online is computationally intensive, we decompose it into three sub-problems (SP1) intra-slice allocation, (SP2) inter-slice allocation, and (SP3) offloading decision and train a distributed hierarchical DeepSets-based model (DeepSets-S) on optimal solutions obtained offline. In the proposed model, we design a slack-aware normalization mechanism for a shared encoder and task-specific decoders, ensuring permutation equivariance over variable-size wireless device (WD) sets. The learned system produces near-optimal allocations with low inference time and maintains permutation equivariance over variable-size device sets. Our experimental results show that DeepSets-S attains high tolerance-based accuracies on SP1/SP2 (Acc1 = 95.26% and 95.67%) and improves multiclass offloading accuracy on SP3 (Acc = 0.7486; binary local/offload Acc = 0.8824). Compared to exact solvers, the proposed approach reduces the execution time by 86.1%, while closely tracking the optimal system cost (within 6.1% in representative regimes). Compared with baseline models, DeepSets-S consistently achieves higher cost ratios and better utilization across COIN/MEC resources.

Distributed Hierarchical Machine Learning for Joint Resource Allocation and Slice Selection in In-Network Edge Systems

TL;DR

The paper tackles the challenge of joint wireless and computing resource management with optimal slice selection in a slice-enabled COIN-MEC edge network for latency-sensitive Metaverse applications. It formulates a mixed-integer nonlinear program (MINLP) and decomposes it into three subproblems (SP1-SP3), then trains a distributed hierarchical DeepSets-S model with a shared encoder and task-specific decoders to approximate solver-derived policies. The model enforces permutation equivariance over variable-size WD sets via a slack-aware normalization, achieving high accuracies (, , , ) and reducing online inference time by , while staying within of the optimal cost. The work demonstrates scalable AI-native orchestration for 6G edge systems and outlines future directions for dynamic environments, semi-supervised learning, and real-world validation.

Abstract

The Metaverse promises immersive, real-time experiences; however, meeting its stringent latency and resource demands remains a major challenge. Conventional optimization techniques struggle to respond effectively under dynamic edge conditions and high user loads. In this study, we explore a slice-enabled in-network edge architecture that combines computing-in-the-network (COIN) with multi-access edge computing (MEC). In addition, we formulate the joint problem of wireless and computing resource management with optimal slice selection as a mixed-integer nonlinear program (MINLP). Because solving this model online is computationally intensive, we decompose it into three sub-problems (SP1) intra-slice allocation, (SP2) inter-slice allocation, and (SP3) offloading decision and train a distributed hierarchical DeepSets-based model (DeepSets-S) on optimal solutions obtained offline. In the proposed model, we design a slack-aware normalization mechanism for a shared encoder and task-specific decoders, ensuring permutation equivariance over variable-size wireless device (WD) sets. The learned system produces near-optimal allocations with low inference time and maintains permutation equivariance over variable-size device sets. Our experimental results show that DeepSets-S attains high tolerance-based accuracies on SP1/SP2 (Acc1 = 95.26% and 95.67%) and improves multiclass offloading accuracy on SP3 (Acc = 0.7486; binary local/offload Acc = 0.8824). Compared to exact solvers, the proposed approach reduces the execution time by 86.1%, while closely tracking the optimal system cost (within 6.1% in representative regimes). Compared with baseline models, DeepSets-S consistently achieves higher cost ratios and better utilization across COIN/MEC resources.

Paper Structure

This paper contains 17 sections, 3 theorems, 61 equations, 9 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related works
  3. Computing in the network (COIN) / In-Network Computing
  4. ML-based Resource Management
  5. System Model
  6. Communication Resources
  7. Computing Resources
  8. Cost Model
  9. Optimization Problem Formulation
  10. Distributed Hierarchical DeepSet-S model for resource management and task offloading
  11. DeepSet Encoder
  12. Decoder
  13. Experiments and Evaluation
  14. Dataset Generation
  15. Experiment Environment
  16. ...and 2 more sections

Key Result

Lemma 1

For every slice $n$ and resource $e$, the allocations of resources per-WD $\{\phi^{n}_{i,e}\}_{i \in \mathcal{I}}$ produced by SP1 satisfy

Figures (9)

  • Figure 1: Slicing enabled COIN-MEC System with $\mathcal{N} = 3$ slices, $\mathcal{I} = 6$ WDs, $\mathcal{A} = 3$ APs, $\mathcal{M} = 1$ MEC and $\mathcal{C} = 6$ COINs
  • Figure 2: Distributed Hierarchical AI Model
  • Figure 3: DeepSets-S Model
  • Figure 4: Absolute error of DeepSet-S model against WD size.
  • Figure 5: Algorithm computation time
  • ...and 4 more figures

Theorems & Definitions (6)

  • Lemma 1
  • proof
  • Lemma 2
  • proof
  • Lemma 3
  • proof