International Journal of Innovative Computer Science and IT Research

E-ISSN: 3067-1108

A Widely Indexed Open Access Peer Reviewed Multidisciplinary Monthly Scholarly International Journal

Call for Paper Volume 2 Issue 6 November-December 2026 Submit your research before last 3 days of this month to publish your research paper in the issue of November-December.

Reinforcement Learning for Autonomous Decision-Making in Dynamic Computing Environments

Author(s) Hector Geffner
Country Germany
Abstract Modern computing environments are increasingly characterized by dynamic workloads, heterogeneous resources, fluctuating network conditions, distributed infrastructures, and continuously changing application requirements. Conventional resource management and decision-making mechanisms generally depend on predefined rules, static optimization models, or manually configured policies, limiting their ability to respond effectively to unpredictable operational conditions. Reinforcement Learning (RL) provides an adaptive computational paradigm in which intelligent agents learn decision policies through repeated interaction with an environment and feedback from observed rewards.
This paper examines the application of reinforcement learning for autonomous decision-making in dynamic computing environments, with particular emphasis on cloud computing, edge computing, distributed systems, resource scheduling, workload management, task offloading, and infrastructure optimization. The study reviews fundamental RL approaches, including Q-learning, Deep Q-Networks (DQN), policy-gradient methods, actor-critic architectures, Proximal Policy Optimization (PPO), and multi-agent reinforcement learning. A conceptual framework is developed in which environmental observations are transformed into states, candidate actions are evaluated through reward functions, and continuously updated policies determine autonomous computing decisions.
The analysis compares representative RL approaches according to adaptability, scalability, computational requirements, decision-making capability, and suitability for dynamic environments. Recent research demonstrates increasing use of deep and multi-agent reinforcement learning for edge-cloud scheduling and resource management, particularly where workload and network conditions vary rapidly. The paper further identifies challenges related to training cost, sample efficiency, safety, interpretability, partial observability, reward design, generalization, and deployment reliability. The findings indicate that RL can provide a strong foundation for self-adaptive computing systems when combined with efficient state representation, robust reward mechanisms, distributed learning, and appropriate safety controls. Future research should focus on explainable, federated, hierarchical, risk-aware, and resource-efficient reinforcement learning architectures capable of operating reliably across heterogeneous computing infrastructures
Keywords Reinforcement Learning, Autonomous Decision-Making, Deep Reinforcement Learning, Dynamic Computing, Cloud Computing, Edge Computing, Resource Management, Intelligent Systems, Task Scheduling, Multi-Agent Learning.
Field Engineering
Published In Volume 1, Issue 3, May-June 2025
Published On 2025-05-12

Share this