Predicting future cellular network traffic volume patterns is crucial for optimizing network resource management and enhancing user experience. Recently, with cellular traffic data represented as pseudo-image data, state-of-the-art frameworks based on Deep Neural Networks (DNNs) have been introduced to enable effective modeling of spatiotemporal dependencies for the prediction tasks. Nonetheless, highly parameterized DNNs require large datasets, which are often missing for cellular network traffic prediction task and thus underscore the need for effective data augmentation strategies. Our preliminary analysis shows that the augmentation methods for natural images prove ineffective in enhancing performance on this task due to its pseudo-image nature. We identify that the reason lies in image augmentations introducing missing values and misalignments, with masking and geometric alterations disrupting the real-world cellular traffic patterns. On top of this, we propose an adaptation of underlying augmentation process into a tailored strategy suitable for the spatial and temporal complexities of the data, which stem from interactions between base stations and evolving traffic patterns. Specifically, we introduce MixScale, an augmentation technique designed for pseudo-image cellular network traffic data. MixScale integrates constrained spatio-temporal data mixing with multi-scaling to better align with the unique characteristics of the dataset. Evaluation on two real-world datasets, including a 5G dataset, demonstrates promising results; MixScale consistently achieves reductions between 16% and 32% in root mean square error (RMSE) compared to the baseline.
ABSTRACT Improving photovoltaics (PV) system performance through simulation requires accurate PV models. The nonlinear relationship between current and voltage, coupled with incomplete manufacturer data, presents a significant challenge in parameter estimation. This research work presents an innovative optimization framework designed to extract the parameters of a triple‐diode model representing an unknown Solar PV module. The proposed approach leverages a novel hybrid optimization technique called the Kookaburra‐based Dwarf Mongoose (KO‐DM) Optimization Algorithm. It determines the unknown parameter that helps to characterize the equivalent circuit of the PV cell described by the triple‐diode model. The performance of the proposed KO‐DM algorithm was assessed using two distinct datasets: (1) 25°C with 1000 W/m² and (2) 45°C with 1000 W/m². The proposed method offers superior performance, exhibiting the ability to efficiently solve complex nonlinear optimization problems, converge rapidly to the global optimum, and involve a simple computational procedure.
Real-time computationally intensive task scheduling for intelligent transportation system (ITS) applications like road safety and traffic forecasting within the deadline while ensuring user quality of experience (QoE) is a complex engineering problem. Meanwhile, adopting Digital Twin (DT) as an emerging technology in vehicular edge computing (VEC) enables efficient capture of real-time state information, thereby addressing the resource scheduling problem in an unpredictable vehicular topology setting. However, exploring strategies to enhance user QoE in timeliness and reliability domains could be a compelling and underexplored research challenge, particularly within the dynamic and trust-sensitive context of vehicular edge computing. In this paper, we have developed an optimization framework using Mixed Integer Linear Programming (MILP), which maximizes user QoE by allocating task execution responsibility to highly reliable and reputed vehicles in a DT-enabled VEC environment. The framework leverages the demand-supply theory of economics to cluster vehicles based on computational resources and applies multi-weighted subjective logic to ensure accurate reputation updates. The NP-hard nature of the formulated optimization problem has driven us to develop an Asynchronous Advantage Actor-Critic (A3C)-based deep reinforcement learning algorithm, namely DARQoE, for offloading tasks in the Internet of Vehicles (IoV). The developed DARQoE framework utilizes effective parallelization across multiple agents with separate environments, accelerating the learning process for IoV task offloading. The experimental results of the developed DARQoE framework demonstrate significant performance improvements in terms of QoE in the timeliness and reliability domains of task execution by up to 15% and 25%, respectively, compared to state-of-the-art works.
The performance of machine learning models in industrial settings is often limited by noise and missing values in real-world data. Tabular data representations, commonly used in traditional machine learning, may not effectively capture complex relationships or maintain reliability under such data degradation. This study comparatively evaluates the robustness of tabular and graph-based data representations for machine learning models when faced with data corruption. Using a real-world steel industry energy consumption dataset, we assess six models: Random Forest, XGBoost, Multi-Layer Perceptron (MLP), Graph Convolutional Network, SAGE, and Graph Attention Network, across clean, noisy, missing, and combined noise and missing data scenarios. A novel transformation technique converts tabular data into graph structures to facilitate relational learning in graph-based models. Graph-based models demonstrated 30.8% greater robustness than tabular models, as measured by their lower average drop in classification accuracy across missing, noisy, and combined data corruption scenarios. These findings pave the way for deploying more resilient artificial intelligence (AI) systems in complex industrial environments, emphasizing the critical role of relational data representations in robust machine learning. For validation, we applied another study with the UCI Machine Learning Repository: the Concrete Compressive Strength Dataset, and found comparable resonance in this regard.
Federated Learning (FL) has become a leading decentralized solution that enables multiple clients to train a model in a collaborative environment without directly sharing raw data, making it suitable for privacy-sensitive applications such as healthcare, finance, and smart systems. As the field continues to evolve, the research field has become more complex and scattered, covering different system designs, training methods, and privacy techniques. This survey is organized around the three core challenges: how the data is distributed, how models are synchronized, and how to defend against attacks. It provides a structured and up-to-date review of FL research from 2023 to 2025, offering a unified taxonomy that categorizes works by data distribution (Horizontal FL, Vertical FL, Federated Transfer Learning, and Personalized FL), training synchronization (synchronous and asynchronous FL), optimization strategies, and threat models (data leakage and poisoning attacks). In particular, we summarize the latest contributions in Vertical FL frameworks for secure multi-party learning, communication-efficient Horizontal FL, and domain-adaptive Federated Transfer Learning. Furthermore, we examine synchronization techniques addressing system heterogeneity, including straggler mitigation in synchronous FL and staleness management in asynchronous FL. The survey covers security threats in FL, such as gradient inversion, membership inference, and poisoning attacks, as well as their defense strategies that include privacy-preserving aggregation and anomaly detection. The paper concludes by outlining unresolved issues and highlighting challenges in handling personalized models, scalability, and real-world adoption.
