Real Time Adaptive AI pipelines for edge cloud systems: Dynamic optimization based on infrastructure feedback

Aravind Chinnaraju *

Seattle, USA.
 
Review
World Journal of Advanced Engineering Technology and Sciences, 2024, 13(02), 887-908 
Article DOI: 10.30574/wjaets.2024.13.2.0636
Publication history: 
Received on 15 November 2024; revised on 28 December 2024; accepted on 30 December 2024
 
Abstract: 
Edge and cloud convergence is reshaping how artificial‑intelligence workloads are deployed, yet most production pipelines remain static and assume stable bandwidth, latency, and power budgets. This article proposes a real‑time adaptive AI pipeline that continuously senses cross‑layer infrastructure telemetry such as bandwidth fluctuation, round‑trip time, packet loss, and thermal headroom, then reconfigures inference and training flows across device, far‑edge, and core‑cloud tiers. A lightweight telemetry bus feeds a reinforcement‑learning control plane that orchestrates split‑model placement, on‑the‑fly quantization, and energy‑aware scheduling while preserving confidential‑compute boundaries. Experiments on a heterogeneous testbed featuring Raspberry Pi 5, Jetson Orin, AWS Graviton, and Nvidia A100 nodes validate that the framework achieves substantial latency and energy improvements under variable 5G and Wi‑Fi 7 backhaul conditions when compared with fixed cloud‑centric baselines. Continuous learning loops further mitigate concept drift by coupling edge data streams directly to training clusters, enabling faster recovery of model accuracy. The design also integrates composite service‑level objectives, AI‑aware chaos engineering, sustainability dashboards, and automated fail‑over orchestration, offering a holistic blueprint for resilient and environmentally conscious AI services. Finally, the paper explores future enablers such as 6G micro‑slicing, neuromorphic coprocessors, and quantum‑assisted route planning, and concludes with a practical adoption roadmap for practitioners and researchers.
 
Keywords: 
Real‑Time Telemetry; Adaptive Inference; Feedback Control Plane; Split‑Model Orchestration; Latency Optimization; Energy‑Adaptive Computing
 
Full text article in PDF: 

Click here

 
Copyright information: 
Copyright © 2024 Author(s) retain the copyright of this article. This article is published under the terms of the Creative Commons Attribution Liscense 4.0