Hyper-parameter Optimization for Wireless Network Traffic Prediction Models with A Novel Meta-Learning Framework

TL;DR

The paper addresses the computational burden of hyper-parameter optimization for large-scale wireless NTP models by introducing a meta-learning framework that transfers optimization experience across tasks. It combines an attention-based base-learner (ADNN) with a two-stage meta-learner (KNN to propose candidate hyper-parameters and an AGA with GRN-based screening to finalize them) to achieve faster optimization and improved predictive accuracy. Key contributions include entropy-based meta-feature analysis linking task traits to hyper-parameter preferences, and a scalable meta-learning pipeline that substantially outperforms traditional optimization methods like BO, GA, and PSO while maintaining robustness across base-learners. The approach has practical significance for deploying DL-based NTP in ultra-dense networks, reducing operational expenditure by speeding up hyper-parameter tuning and improving forecast quality.

Abstract

This paper proposes a novel meta-learning based hyper-parameter optimization framework for wireless network traffic prediction (NTP) models. The primary objective is to accumulate and leverage the acquired hyper-parameter optimization experience, enabling the rapid determination of optimal hyper-parameters for new tasks. In this paper, an attention-based deep neural network (ADNN) is employed as the base-learner to address specific NTP tasks. The meta-learner is an innovative framework that integrates meta-learning with the k-nearest neighbor algorithm (KNN), genetic algorithm (GA), and gated residual network (GRN). Specifically, KNN is utilized to identify a set of candidate hyper-parameter selection strategies for a new task, which then serves as the initial population for GA, while a GRN-based chromosome screening module accelerates the validation of offspring chromosomes, ultimately determining the optimal hyper-parameters. Experimental results demonstrate that, compared to traditional methods such as Bayesian optimization (BO), GA, and particle swarm optimization (PSO), the proposed framework determines optimal hyper-parameters more rapidly, significantly reduces optimization time, and enhances the performance of the base-learner. It achieves an optimal balance between optimization efficiency and prediction accuracy.

Figures (15)

• Figure 1: The mean square error (MSE) performance of prediction models with different hyper-parameter selection strategies for three randomly selected mobile cells: a) mobile cell 1595; b) mobile cell 2535; c) mobile cell 3040.
• Figure 2: The information entropy and the conditional entropy (CE) of the best values of each kind of hyper-parameters.
• Figure 3: The proposed hyper-parameter optimization framework for cell-level wireless NTP model.
• Figure 4: The structure of the base-learner.
• Figure 5: The schematic diagram of the Multi-Head Self-Attention mechanism.