Zero Touch Predictive Orchestration: Automating Time-Series Models for the Cloud-Edge Continuum

Noosaga Placements

• Machine Learningsubfield95%
  The paper focuses on automated time-series forecasting using Neural Architecture Search, a core machine learning technique, and evaluates models on standard regression metrics (MSE, MAE, MAPE). The primary arXiv category is cs.LG.
  we propose a fully automated time-series prediction architecture driven by a novel data-mixing methodologyProcessed through a Neural Architecture Search (NAS) engine, the system automatically generates highly accurate baseline modelsExperimental results demonstrate that merging the target data with TimeTrack effectively mitigates the cold start challenge
• Recurrent Neural Networksframework60%
  The NAS search space includes Recurrent Neural Networks (RNN, LSTM, GRU) which are used as candidate architectures for time-series forecasting, but the paper does not extend or critique RNNs specifically.
  The architectural search space ... encompassing ... standard Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRU)
• Network Architecturesubfield70%
  The paper addresses resource orchestration in the Cloud-Edge Continuum, a network architecture concept, and introduces a Resource Exposer for dynamic node discovery and telemetry collection in distributed edge networks.
  The Cloud-Edge Continuum (CEC) enables latency-critical applications by distributing resources to the far edgewe introduce a lightweight, technology-agnostic Resource Exposer (RE) that dynamically discovers nodes and continuously collects customizable telemetryThis decentralized method allows the system to scale flexibly, tracking nodes as they dynamically join or leave the network
• Statistical Learningframework50%
  The paper uses standard machine learning evaluation metrics (MSE, MAE, MAPE) and trains predictive models, which situates it within the statistical learning framework, but it does not engage with statistical learning theory directly.
  forecasting accuracy measured in Mean Squared Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE)
• Deep Learningsubfield60%
  The paper uses deep learning architectures (LSTM, GRU, CNN, RNN, TCN, Transformer) within the NAS search space for time-series forecasting, but the focus is on automation and data mixing rather than advancing deep learning theory.
  The architectural search space ... encompassing Multi-Layer Perceptrons (MLP), standard Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRU), Convolutional Neural Networks (CNN), Temporal Convolutional Networks (TCN), and Transformersthe prevailing approach for training Deep Neural Networks (DNNs)—including time-series models—remains a manual, trial-and-error process
• Convolutional Neural Networksframework50%
  CNNs are included in the NAS search space as a candidate architecture for time-series forecasting, but the paper does not focus on CNNs specifically.
  The architectural search space ... encompassing ... Convolutional Neural Networks (CNN)
• Transformer Architectureframework50%
  Transformers are included in the NAS search space as a candidate architecture, but the paper does not focus on the Transformer architecture specifically.
  The architectural search space ... encompassing ... Transformers

Abstract

The Cloud-Edge Continuum (CEC) enables latency-critical applications by distributing resources to the far edge, but its extreme volatility makes proactive Zero Touch Management via time-series forecasting essential. However, orchestrators face a severe "cold start" problem: newly discovered nodes lack the historical data required to train localized predictive models, while generalized models fail to capture unique hardware and microservice behaviors. To solve this, we propose a fully automated time-series prediction architecture driven by a novel data-mixing methodology. At the infrastructure level, we introduce a lightweight, technology-agnostic Resource Exposer (RE) that dynamically discovers nodes and continuously collects customizable telemetry (e.g., compute, network, energy). To overcome the sparsity of these initial local samples, our framework automatically merges them with TimeTrack, our publicly available, high-resolution dataset collected at 45-second intervals. This synergizes TimeTrack's foundational, high-frequency temporal patterns with the precise calibration of the local node data. Processed through a Neural Architecture Search (NAS) engine, the system automatically generates highly accurate baseline models. Experimental results demonstrate that merging the target data with TimeTrack effectively mitigates the cold start challenge. This integration significantly improves forecasting accuracy measured in Mean Squared Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) and accelerates convergence compared to training on the sparse local samples alone, training solely on generic datasets, or mixing the target data with standard alternative datasets, establishing a robust foundation for continuous MLOps deployment.

Paper Context