کنفرانس بین المللی فناوری اطلاعات و دانش

English

صفحه اصلی / شانزدهمین کنفرانس بین المللی فناوری اطلاعات و دانش

Comparative Study of Deep Reinforcement Learning and Genetic Algorithm Approaches for IoT Machine Learning Job Deployment in Fog Computing

نویسندگان :

Amir Moazeni¹ Omid Bushehrian²

1- دانشگاه صنعتی شیراز 2- دانشگاه صنعتی شیراز

کلمات کلیدی :

Smart City،Cloud-Edge Computing،IoT Task Scheduling،DDPG،MILP،Federated Learning

چکیده :

The growing adoption of federated learning (FL) in Internet of Things (IoT) applications demands efficient and scalable task deployment across heterogeneous cloud–fog infrastructures. This work conducts a unified comparative study of six representative approaches—Deep Deterministic Policy Gradient (DDPG), DDPG Greedy, Genetic Algorithm (GA), Integer Linear Programming (ILP), Edge IoT, and Cloud IoT—classified into online and offline optimization methods. The evaluation is performed on an FL application using an LSTM model trained over multi site meteorological datasets, measuring deployment latency, execution cost, and adaptability under dynamic workloads. Results show that DDPG achieves an effective trade off, reducing cost by up to 18% compared to GA while maintaining sub second placement latency. ILP yields the lowest overall cost but lacks adaptability for real time environments, whereas heuristic baselines reveal the latency–cost balance between decentralized edge processing and centralized cloud execution. The findings provide quantitative guidance for selecting task deployment strategies in large scale smart city FL scenarios.