The increasing demand for real-time carbon emission monitoring in distributed edge environments has highlighted the need for accurate prediction models that consider spatial and temporal context factors. Traditional centralized prediction models face significant challenges in processing large-scale emission data in real-time, especially when dealing with fluctuating data from diverse locations and time intervals. These challenges are further compounded by the need for privacy protection when handling sensitive environmental data. To address these issues, this paper presents a distributed approach to carbon emission prediction in edge environments, incorporating spatial-temporal context factors such as location and time of day. By leveraging locality-sensitive hashing (LSH) techniques, we propose a privacy-preserving model that enables efficient prediction across edge nodes while maintaining user privacy. Experimental results demonstrate the feasibility and effectiveness of the proposed method in accurately predicting carbon emissions, with a significant reduction in computational overhead compared to centralized models. This work provides a practical solution for real-time carbon emission prediction and privacy-preserving environmental monitoring in distributed systems.

Modern broadband access networks increasingly carry heterogeneous, delay-sensitive traffic generated by cloud-assisted applications, real-time communication services, and short-form multimedia exchanges. These workloads have bursty ON/OFF packet dynamics, creating dense arrival clusters that conventional GPON/XGS-PON upstream schedulers struggle to handle. This work proposes an AI-driven Dynamic Bandwidth Allocation (AI-DBA) framework based on a Deep Q-Network (DQN) that observes global queue states for video, voice, and data traffic and performs adaptive admission/prioritization/scheduling decisions. The DQN agent learns an adaptive scheduling policy through experience replay and reward-driven interactions to anticipate burst formations, drain high-pressure queues, and smooth resource allocation. Simulation results show that DQN-DBA maintains near-zero buffer queues and eliminates observed packet loss across all traffic classes, and stablize the delay and jitter profiles, whereas TCP and QCT-ARED experience long queue buildup, increased loss rates, and high delay/jitter variability under the same bursty traffic conditions. Across both experiments, the proposed approach consistently outperforms all baselines. Against the Non-AI scheme, the AI-based model reduces packet loss by 72.84% and average waiting time (latency) by 41.73%. In the second experiment, across video, voice, and data traffic, DQN-DBA achieves a 100% reduction in observed packet loss compared to both TCP and QCT-ARED, reduces steady-state delay by approximately 74.7% relative to TCP and 66.3% relative to QCT-ARED, and improves throughput by about 558–634% over TCP and 343–384% over QCT-ARED.The results confirm that AI-driven upstream scheduling provides a robust, scalable, and highly adaptive solution for managing bursty multi-traffic loads in next-generation PON architectures.

With the rapid development of cloud-based creative platforms, personalized visual art recommendation has attracted increasing attention in recent years. Online illustration services usually host a large number of artworks with diverse styles and themes, making it difficult for users to efficiently discover content that matches their visual preferences. Existing recommendation methods mainly emphasize preference matching at the individual artwork level, while the overall quality of experience (QoE) perceived from a recommended artwork set is often insufficiently considered. In this paper, we propose a QoE-driven cloud intelligence approach for personalized visual art recommendation. Based on historical user interaction data, a weighted artwork correlation graph is constructed to characterize perceptual relationships among artworks. By jointly considering user intent coverage and cross-artwork perceptual consistency, a QoE-aware optimization strategy is designed to generate a compact and coherent recommendation list. Experimental results on a large-scale dataset show that the proposed approach achieves better performance than several baseline methods in terms of recommendation accuracy and QoE-related metrics, while maintaining reasonable computational efficiency in cloud environments. These results indicate that incorporating QoE considerations into cloud-based visual art recommendation can improve user experience and service effectiveness.

Code summarization is an important task in software engineering that helps developers understand and maintain code by generating natural language summaries. Existing approaches predominantly rely on single models, facing a dilemma: directly deploying large language models (LLMs) incurs high training costs, while lightweight models specialized for summarization are constrained by the quality of training data and their ability to capture the complex structural semantics of code. This highlights the urgent need for synergistic collaboration between large and small models in cloud computing environments. To address these issues, this paper proposes a cloud-assisted code summarization framework. First, we achieve code enhancement by invoking cloud-deployed LLM services. The specific workflow involves using preset prompt templates to guide the model in evaluating code quality and automatically repairing defects based on its feedback, thereby constructing high-quality datasets Java-QE and Python-QE. Second, for efficient edge deployment, we introduce HiSum: AST Hierarchy-Aware Code Summarization model, a lightweight model. HiSum transforms code AST into Directed Syntax Graphs (DSG) to preserve structural semantics, encodes them via a directed graph convolutional network and decode to improve summary quality. Experimental results show that our framework significantly enhances code summarization performance. On the constructed Java-QE and Python-QE datasets, the HiSum model achieves notable improvements over state-of-the-art baselines in BLEU, METEOR, and ROUGE-L metrics (increases of 1.06%, 1.98%, 3.12% for Java-QE, and 1.46%, 3.24%, 2.20% for Python-QE, respectively). This research provides a solution that utilizes cloud LLM-assisted data enhancement to empower a lightweight hierarchical-aware model.

Vehicular mobile crowdsensing (VMCS) has emerged as powerful and promising paradigm for mobile crowdsensing (MCS) by deploying sensing devices on connected vehicles to collect perception data. Task assignment plays a crucial role in determining the efficiency and performance of in VMCS systems. However, the large number of connected vehicles and sensing tasks involved, coupled with the implication of task information and vehicle status over time, which poses new challenges to system stability. Therefore, this paper puts forth an edge computing-enabled task assignment for VMCS, seeking to enhance the efficiency of task assignment by scheduling a reasonable match between sensing tasks and connected vehicles, and to reduce decision-making delays. Specifically, the task assignment problem is acknowledged as an optimization process with the objective of reducing the tasks waiting duration by considering the task completion sequence rationality under the task time constraint and the energy limitation of the vehicles, and is demonstrated to be an NP-hard problem. Firstly, a valid task sequence is mainly generated for vehicles prior to task assignment to reduce the collision rate between different lists of vehicular tasks. Secondly, this issue is resolved by employing a cooperative game theory-based coalition formation game. On the basis of the introduction of the coalition formation game, the switching rule and historical selection set are used to prevent cyclical coalition switching, and a convergence-guaranteed algorithm with low complexity is proposed to achieve a Nash stable solution. Finally, a thorough set of experiments are performed under various dynamic conditions to show that our suggested scheme is superior to other comparison schemes in terms of task completion percentage, average waiting time, and running time.

The effective maintenance of Mobile Applications (MAPPs) relies on the timely and accurate triage of bug reports, where severity labels determine the order of issue resolution. Manual severity assessment is often labor-intensive, error-prone, and inconsistent, highlighting the need for automated approaches that capture both semantic and contextual information. This study introduces RAG-GPT-SBR, a Retrieval-Augmented Generation framework that leverages a fine-tuned GPT-2 model for automated bug severity prediction in Cloud-based MAPPs. In addition to textual content, the framework incorporates non-textual metadata such as device type, operating system, crash signature, and bug frequency. Each new report is enriched with semantically related historical cases retrieved from a curated knowledge base using a hybrid BM25 and dense retrieval mechanism, forming structured input prompts for GPT-2. The fine-tuned model encodes this composite information and generates contextual embeddings, which are fed into a classification head to predict severity levels. Experiments conducted on a public Hugging Face bug report dataset demonstrate that RAG-GPT-SBR outperforms existing deep learning baselines, achieving gains of 7.98%, 8.20%, 7.90%, and 8.10% in accuracy, precision, recall, and F1-score, respectively. The results affirm that combining retrieval-augmented context with generative language models enhances reliability, scalability, and consistency in automated mobile application maintenance.

The synergy between powerful, cloud-hosted large models and efficient, on-device small models presents a new frontier for mobile artificial intelligence. However, enabling real-time, knowledge-intensive interactions, such as querying physical objects in the user’s environment, is severely hampered by the high latency of cloud-centric approaches and the limited reasoning capabilities of edge-only models. This paper introduces a novel synergistic cloud-edge intelligence framework designed to bridge this critical gap. Our framework implements a semantic partitioning of labor: a lightweight edge client performs real-time perception to generate compact multimodal features, while an intelligent trigger decides when to offload a structured, minimal-data payload to a powerful cloud-based reasoning engine for definitive entity linking and knowledge-grounded question answering. We validate our approach on SynergyQA, a new benchmark for this task. Experimental results show that our framework reduces end-to-end latency by over 80% and data transmission by over 98% compared to naive offloading strategies, while achieving question-answering performance that is competitive with state-of-the-art cloud-only models. Our work provides an efficient and robust blueprint for deploying the next generation of interactive, context-aware AI on mobile and edge devices.

The rapid growth of fog-supported Internet of Things (IoT) networks presents significant challenges in securing efficient and scalable communication between heterogeneous devices. The age-old authentication and key management schemes like RSA and hash are prone to emergent attacks and are characterized by high computational cost. To address these challenges, the paper will present a lightweight Elliptic Curve Cryptography (ECC)-based mutual authentication and key management system, combined with the Advanced Encryption Standard (AES), to ensure the safety of data transmission. The suggested model uses a Gateway Node (GWN) as a semi-trusted intermediary to enhance the process of registration of devices, verifying their identities, and determining a session key between the IoT devices and fog nodes. Formal security analysis and mathematical formulations prove that the scheme has mutual authentication, freshness of keys, forward secrecy, and resistance to replay and impersonation attacks. Performance analysis based on NS-3 (v3.40) reveals that the proposed framework has 48% reduced latency and 35% lower cost of computation than the traditional RSA-based models, while also having good security properties. The findings affirm that the proposed ECC-AES model provides an efficient trade-off between security and computation efficiency, and can be used in real-time and resource-constrained fog-IoT systems.