Jul 2026· 2026 International Conference on Intelligent and Sustainable AI Systems (ICOSAAS)· pp. 143-149· 0 citations· 26 references
Abstract
Generative AI (GAI) refers to advanced models that can generate new, realistic, context-aware data or solutions by learning from existing datasets, making them extremely valuable in adaptive intelligent systems. Traditional AI technologies often face limitations, including a lack of interpretability, sensitivity to noisy data, and an inability to generalize to dynamic environments. These limitations can be effectively addressed by GAI-driven adaptive neuro-fuzzy inference systems (ANFIS). To facilitate the ease of scalable processing and real-time deployment, the proposed framework is developed in a cloud computing system, whereby the aggregation of large-scale QoE data, the training of generative models and the optimization of the fuzzy rules are handled using the cloud computing resources, and latencysensitive inference is done effectively at the edge. To address these challenges, we propose a Generative Reinforcement Learning (GRL) method that enhances decision-making skills by generating artificial experiences, utilizing the video transmission system as an environment, and learns policies to optimize latency. The Generative Diffusion Model (GDM) for feature denoising offers a powerful mechanism for removing noise in high- dimensional data, thereby enhancing the accuracy and stability of prediction tasks. Physics-Guided Generative Modeling (PG2M) combines domain-specific physics laws with AI learning to ensure scientifically consistent and interpretable outputs. Finally, Generative Adversarial Networks (GANs) are employed for generative rule evolution, combining evolutionary computation with adversarial learning to dynamically generate and improve classification rules. ANFIS utilizes fuzzy rules to model delay shapes, while GRL optimizes video transmission strategies based on the output of ANFIS. Reduce latency dynamically by learning adaptive frame scheduling; it improves network responsiveness to fluctuations. The presented method achieved 94% accuracy and improved the latency of the video communication service.
Computation methods of artificial intelligence are gradually shifting from cloud computing to edge computing and on-device machine learning (ODML). How to contribute an effective machine learning model in the resource-limited environment, has become a significant and rapidly evolving research field. The training and inference of deep learning model used to be performed on the cloud high-performance computing clusters. There are many problems with uploading data to cloud, for example, high latency, round-trip latency, security issues, and a lack of privacy guarantees, and in this case, people cannot make real-time decisions. So, using edge devices to process tasks can significantly decrease the cost of transmission. The need of low latency, quick response, privacy protection and high adaptability has become the drive force of this change. This report aims to provide a comprehensive overview of lightweight model inference technologies in edge computing environments, mainly targeting low performance devices, such as mobile phones, intelligent equipment in vehicle, VR/AR headsets and Internet of Things (IoT). This paper introduces efficient learning and inference on edge devices from four aspects: 1) the definition of core terminology and concrete application environment; 2) the core technology of model compression, neural networks and knowledge distillation, which is used to deal with the tasks in the resource-limited environment; 3) the standards for evaluation of time/space complexity; 4) the challenge and opportunity which people face currently and in future.
Hengrui Bi· Applied and Computational En...· 0 citations
With the advent of seventh-generation (7G) wireless systems, the spectrum environment is extremely dynamic and heterogeneous, and traditional methods of sensing do not offer reliable and efficient performance. This paper introduces a self-evolving spectrum sensing system to enable adaptive and intelligent spectrum access based on a deep reinforcement-based learning paradigm. The framework depicts the sensing process as a sequence decision problem where an autonomous agent is able to continually refine its policy as it engages with the environment. The multi-objective reward formulation is designed to maximize the combination of the detection accuracy, false alarms, energy usage, and using the spectrum. Moreover, an adaptive representation of state mechanism is also introduced to indicate the temporal changes and short-term changes in the spectrum occupancy. The self-evolution strategy proposed adjusts learning parameters and decision policies in a dynamic manner that gives a robust operating in a non-stationary environment. The overall analysis of the experiment shows that the structure achieves detection probability of 97.1, false alarm rate reduced to minimum of 3.8, spectral usage maximized to above 92 and convergence rate is quicker as compared to the existing techniques. These results confirm the appropriateness of the proposed method in overcoming the issues of the next-generation wireless systems.
A.L Sriram, H. N. Divya, S. Sabarinathan et al.· International Conference on...· 0 citations
Reinforcement Learning (RL) has shown remarkable success in enabling adaptive and data-driven optimization for various applications in wireless networks. However, classical RL suffers from limitations in generalization, learning feedback, interpretability, and sample efficiency in dynamic wireless environments. Large Language Models (LLMs) have emerged as a transformative Artificial Intelligence (AI) paradigm with exceptional capabilities in knowledge generalization, contextual reasoning, and interactive generation, which have demonstrated strong potential to enhance classical RL. This paper serves as a comprehensive tutorial on LLM-enhanced RL for wireless networks. We propose a taxonomy to categorize the roles of LLMs into four critical functions: state perceiver, reward designer, decision-maker, and generator. Then, we review existing studies exploring how each role of LLMs enhances different stages of the RL pipeline. Moreover, we provide a series of case studies to illustrate how to design and apply LLM-enhanced RL in low-altitude economy networking, vehicular networks, and space–air–ground integrated networks. Finally, we conclude with a discussion on potential future directions for LLM-enhanced RL and offer insights into its future development in wireless networks.
Traffic collisions and congestion represent significant challenges within intelligent transportation systems (ITS). Consequently, a vehicular ad-hoc network (VANET) has been established. Numerous architectures have been incorporated into VANETs to manage the extensive data generated by vehicles. Collaboration with fog computing is vital, particularly for applications requiring real-time processing. In addition, there is a growing demand for advanced intrusion detection methods. These techniques are employed to identify the optimal response that the fog server should provide based on data received from a vehicle. As the network continues to grow, the amount of data needing analysis also increases. Thus, deep learning approaches become increasingly efficient. The requirement for feature selection is reduced when leveraging deep learning techniques. This work uses two deep learning-based misbehavior classification schemes for intrusion detection in VANETs: Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN). The vehicular data that the Roadside Units (RSUs) obtain is initially sent to the fog server for preprocessing and classification. This study suggests four classification methods, which include 4 LSTM, 2 LSTM-2 CNN, 3 CNN-1 LSTM, and 2 CNN-1 LSTM within two main classifiers. The first classifier uses them to classify each type of vehicle, while the second uses them to classify generic behavior categories. The results of the first classifier were higher than those of the existing work. The precision ranges from 94% to 98%. The recall ranges from 92% to 97%. The F1 scores range from 94% to 97%. The second classifier also scores the best results compared to other works. The precision, recall, and F1-score range from 99% to 99.66%. The time and space complexity are calculated for each model in each classifier.
H. Kamal, A. Haikal, Mahmoud M. Saafan· Scientific Reports· 0 citations
The recent proliferation of artificial intelligence (AI) applications in mission-critical and latency-sensitive domains, such as autonomous driving, remote healthcare services, smart manufacturing, and immersive extended reality, puts forward requirements on intelligent yet low-latency and scalable computational infrastructures. The underlying reason is that such a transformational paradigm can reduce or circumvent the increasing limitations of traditional cloud-based AI processing, since high communication latency, bandwidth bottlenecks, and lack of context awareness are the bottleneck issues holding back conventional AI processing nowadays. The motivation for such a shift is also advanced by the expectations for sixth-generation (6G) wireless network-based solutions with ultrareliable low-latency communication, integrated sensing and communication, intelligent reflecting surfaces, and native support for AI-native protocols. This survey provides a systematic and critical review of the current status of distributed edge intelligence and focuses on the ability to enable efficient AI inference over 6G wireless networks. The state-of-the-art is categorized across system architectures, distributed learning frameworks, MAC/RAN co-design, resource optimization strategies, and AI computation offloading. Key challenges are identified, including heterogeneity, energy efficiency, context awareness, and service continuity. Furthermore, the most promising future research directions are outlined to achieve fully autonomous, intelligent, and scalable 6G-edge ecosystems able to provide real-time AI services.
B. Vijay, M. Varshini, R. Chaithra· AI Computer Science and Robo...· 0 citations
Video conferencing systems are essential in supporting smooth long-distance cooperation where there should be minimal delays and effective use of computational and network resources. The traditional resource allocation algorithms are not able to change according to the dynamic network conditions, session time, and the variability of the device capacities and lead to contention, latency, and low Quality of Service (QoS). To cope with these issues, the proposed research is an AI-based Sugeno Fuzzy Inference System (SFIS) which is suggested to be used in flexible scheduling of tasks and distribution of resources. The model is trained and tested on the Web Camera People Behavior Dataset which is a collection of session logs, network statistics, CPU and memory utilization, and user activity logs giving a detailed system history. Preprocessing is also performed with Winsorization to help reduce the effects of the extreme outliers and Recursive Feature Elimination (RFE) is used to determine the most significant features. The scheduling model based on SFIS has good predictive performance Accuracy of 94.67%, Precision of 94.00%, Recall of 95.50%, and F1 Score of 94.74%. The analysis of the CPU and memory consumption, latency alleviation, throughput and the rate of accomplishing the tasks shows that there is a great improvement compared to the traditional strategies.
