Tools
Tools
Khulbe, Om, Muntean, Cristina Hava and Gupta, Shaguna (2026) A Multi-Layer Framework for Adaptive Resource Allocation in Edge Computing: A Hybrid Reinforcement Learning Approach. In: 2026 29th Conference on Innovation in Clouds, Internet and Networks (ICIN). IEEE, Athens, Greece. ISBN 979-8-3315-8382-8
Full text not available from this repository.Abstract
In edge computing systems, many devices compete due to limited resources, and Reinforcement Learning (RL) has recently been introduced to solve this problem and share resources. Existing incentive mechanisms that are commonly used across edge computing systems, such as an auction mechanism, where devices bid for resources, and a reputation mechanism, where devices that are more trustworthy are given priority. Most of the research focuses on making devices bid smarter in an auction or using isolated strategies when competing for resources. Some research also implements advanced RL techniques such as Deep Q-Learning (DQN) or Deep Deterministic Policy Gradient (DDPG). The key observation is that when reputation mechanisms are combined with RL, devices are treated as add-ons, and end-to-end RL is rarely used to improve the system. This research study focuses on building an edge computing solution where RL helps to learn the best rules for rewards, trust, and fairness, and not just using fixed rules or helping agents. This work introduces a two-layer hybrid reinforcement learning framework that combines DQN-based device bidding with a DDPG-driven system agent capable of dynamically adjusting pricing and fairness rules. Across three evaluation scenarios, the framework consistently improves fairness, utilization, and revenue stability, particularly under volatile market conditions. The results show that fully adaptive incentive mechanisms can outperform static and semi-dynamic baselines, demonstrating the potential of end-to-end RL to enhance resource allocation in Mobile Edge Computing environments.
Actions (login required)
 |
View Item |