A Semantic-Aware Data Offloading Approach for Latency Reduction in Cloud-Based IoT Environments

Abstract

Now a days , the escalation of Internet of Things (IoT) applications, the volume of data that is being generated is expandingly placing substantial demands in the areas of latency, bandwidth consumption, and computational resources. The traditional cloud-based offloading methods are usually characterized by large delays in communications, thus not accommodating applications that are latency sensitive and in real-time. To overcome these drawbacks, the paper will present a new semantic-sensitive hybrid data offloading system in the cloud-based IoT setting. The suggested solution incorporates three aspects hybrid CNN-Transformer-based semantic segmentation and adaptive feature encoding and reinforcement learning-based edge and cloud collaboration. A multi-agent reinforcement learning model is utilized to dynamically optimize the offloading decisions with respect to network conditions and computation demands and the requirements of the application. Also, a feedback mechanism based on the concept of a digital twin is introduced to allow adaptation of the system in real-time and optimization of its performance. Comprehensive experimental analysis has shown that the offered technique can reduce the latency by 50 percent, make the energy consumption much more efficient by 40, and increase the accuracy to 94.5 percent compared to the traditional CNN-based, Transformer-based, and traditional edge/cloud-only techniques. Besides, the suggested framework is always at a high Quality of Service (QoS) and close to a Pareto-optimal balance of latency and energy consumption. The findings support the legitimacy of the suggested method to deal with the critical issues of contemporary IoT systems. This paper offers a scalable, adaptable and efficient solution to next-generation IoT applications, such as smart cities, autonomous systems and the automation of industries to bring about intelligent edge/cloud collaborative computing.