Vol. 45 · No. 3

This paper deals with the selection of investment portfolios with cardinality constraints in the context of the Markowitz model. To this end, a multiobjective memetic algorithm (MA) featuring a novel hybrid combination of population variation based on elite memory and local search is introduced. These heuristic add-ons revolve around the use of the Sharpe index, a return/risk measure used to evaluate the efficiency of the portfolios generated, allowing to focus the search on relevant areas in the nondominated front. The elite memory exploits this indicator to maintain a record of the most effective solutions found and is used as a tunable intensification mechanism. A sensitivity analysis of the algorithm to determine a suitable parameterization is first performed, and then a comparison is made with seven multiobjective evolutionary algorithms. Data from the equity financial market of the Colombian Stock Exchange between 2010 and 2016 are taken as a benchmark and confirm the great potential of the MA for this problem.

Change detection (CD) in remote sensing (RS) images has become an essential tool for land resource planning, disaster monitoring and urban growth analysis. Despite significant progress with deep learning (DL) methods, traditional approaches struggle to effectively extract multi-scale features and balance local and global information while maintaining computational efficiency. This research introduces NAViT-UNet a novel framework leveraging Neighbourhood Attention Vision Transformers (NAViT) and Multi-Temporal Fusion (MTF) Networks for accurate and efficient CD in multi-temporal RS imagery. The proposed model integrates NAViT modules for local feature extraction, reducing computational overhead while enhancing performance. A robust Multi-Temporal (MT) Fusion Network dynamically aggregates temporal feature maps, optimizing the utilization of complementary MT information. The bottleneck layer incorporates Cascaded Atrous Spatial Pyramid Pooling (ASPP) to broaden the feature receptive field, enabling the precise segmentation of small and complex regions. To address imbalanced segmentation challenges, a Mixed Loss Function combining MS-SSIM Loss, Tversky Loss and Focal Loss ensures fine-grained boundary delineation and segmentation accuracy. The model also adopts the Swish activation function which enhances performance over traditional ReLU. Experiments demonstrated the proposed system achieved best performance in terms of 99.8

This study proposes a novel optimization strategy that integrates the combinatorial assignment technique with the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS), enhancing management efficiency in smart manufacturing. The combinatorial assignment method effectively addresses challenges in decision-making by integrating subjective and objective weight determination, ensuring a more balanced and rational weight distribution. Concurrently, the TOPSIS method facilitates ranking and selecting optimal solutions. Leveraging this integrated approach, the strategy efficiently evaluates employee performance, a critical component of smart manufacturing management. Results demonstrate that this method improves assessment accuracy and fairness, contributing to more effective management practices in the rapidly evolving manufacturing landscape.

Diabetic retinopathy (DR) is the leading cause of vision loss in the world, particularly in individuals who have had a 10 or 15 years history of diabetes. Early detection of the disease and automated fundus image analysis techniques would decrease the possibility of vision loss. In this study, a transfer learning method is proposed, which includes a Convolutional Block Attention Module (CBAM) and optimizes the DenseNet121 architecture, resulting in better feature representations, as well as the focus on relevant retinal area discoveries. The model was trained and evaluated using the publicly available Kaggle Diabetic Retinopathy dataset, which contains labeled color fundus images spanning multiple DR severity levels. To ensure a comprehensive evaluation, metrics such as accuracy, precision, recall, and F1-score were employed. We have obtained 97

The increasing adoption of NoSQL databases to manage diverse and complex data structures necessitates robust modeling techniques. The Unified Modeling Language (UML), traditionally utilized for relational databases, has emerged as a promising tool for modeling NoSQL databases, despite their flexible and schema less nature. This paper presents a comprehensive exploration of UML's role in NoSQL data modeling through a systematic literature review (SLR). Key contributions include an extensive bibliometric analysis of UML applications across various types of NoSQL databases, as well as the identification of trends and gaps in current methodologies. The discussion highlights UML's strengths in the initial stages of conceptualizing data structures while addressing its limitations in representing the dynamic and distributed characteristics inherent to NoSQL systems. Novel insights are provided into the adaptation of UML for logical and physical modeling in document-oriented, graph-based, and columnar databases. The work concludes with actionable recommendations for enhancing UML-based methodologies and outlines future research directions to tackle unresolved challenges in this rapidly evolving field.

Iterative learning is based on the continuous iterative updating of model parameters, which does not need an accurate system model, and can realize the control of dynamic systems with high uncertainty characteristics in a limited time. It is widely used in many fields, such as robots, UAVs, transportation, networks, finance and medical treatment. The distributed multi-agent based on iterative learning distributes large-scale data on multiple nodes for parallel processing, which speeds up the learning speed and improves the robustness of the algorithm. However, the multi-agent system depends on the accurate information transmission between individuals to complete the target task. Due to the complexity of the environment and uncertain factors, such as data loss, delay, noise and other interference, it must be considered. In this paper, an iterative learning control protocol based on local data and a D-type iterative learning algorithm are designed for nonlinear time-varying multi-agent systems with communication interference under the premise that the initial state can be reset to zero. Simulation results show that the proposed algorithm can effectively solve the consistency problem and suppress any type of disturbance, which has strong theoretical and practical significance.

Due to the limitations of device resources and the dynamic nature of data in edge computing environments, developers face significant challenges in acquiring, processing, and analyzing high-quality streaming data during development and testing. This paper examines the characteristics of streaming data in edge computing environments, summarizes its volatility and changing trends in the real world, and deduces potential periodic patterns inherent in such data. A model for generating realistic streaming data in simulated edge computing environments is proposed to address these challenges. The model first preprocesses the original streaming data to extract time-related information (timestamps or precise time). Then, the NSA algorithm is applied to reduce the data volume, followed by the use of the PSD parallel algorithm to transmit the sampled data. Experimental comparisons demonstrate that the model effectively preserves the volatility and trend characteristics of the original streaming data, providing a viable solution for the difficulties developers face in obtaining and processing high-quality streaming data in edge computing environments during development and testing.

Exploitation of the Linux kernel continues to pose significant security risks from sensitive information leakage to severe privilege escalation. While developers prioritize reducing primitives to limit exploitability, kernel heap exploitation remains a persistent threat that significantly threatens kernel integrity. Discovering such primitives is commonly considered an art; existing approaches rely on unsound static analysis and ad-hoc expertise, depending on expertise and limiting practicality in the real-world Linux kernel. In this paper, we present , a systematic approach to automatically discover heap layout manipulation (HLM) primitives in recent kernel heap vulnerabilities, with a design that can scale to edge environments for autonomous, real-time security. The key innovation of lies in combining dynamic analysis via fuzzing with static instrumentation, enabling a direct correlation between primitives and fuzzing seeds. We utilize object-driven instrumentation with a restricted mutation-based fuzzer targeting specific object or cache interactions, such as allocation, deallocation and usage. By enhancing for precise tracking, we lay a foundation that enables the identification and generation of exploitation paths without side effects, ensuring that the discovered primitives are both reliable and practical. To demonstrate effectiveness, we evaluated on 40 real-world CVEs and 240 objects in the stable Linux kernel tree. Our results show that discovers 1.6x more primitives and identifies 4.1x more sensitive objects compared to state-of-the-art approaches.

Open Language Models (OLMs) and edge intelligence have become pivotal in Natural Language Processing (NLP), showcasing significant potential in language generation, text classification, machine translation, and human–computer interaction. This paper focuses on ChatGPT, an open-source large dialogue model based on OpenAI's GPT architecture. It is an emerging trend in the field of artificial intelligence, which aims to expand the language model capabilities on powerful central servers to edge devices, thereby achieving more efficient and flexible applications. Utilizing a Transformer-based encoder–decoder framework with multi-head self-attention mechanisms, ChatGPT excels in generating natural, contextually-aware language responses, akin to human conversation. Edge intelligence combined with ChatGPT can provide more intelligent and personalized services. Despite its advancements, challenges remain regarding model complexity and data demands. We propose enhancements in training methodologies, model architecture, and application domains to foster more intelligent and fluid dialogue systems. Our research highlights ChatGPT and edge intelligence's contributions to both academic inquiry and practical innovation, positioning it to deliver high-quality AI services across various fields.

With the development of smart cities, energy saving has become an important goal, and intelligent streetlight systems (ISLS) play an important role in reducing power consumption. For energy saving purposes, this paper proposes an ISLS based on LoRaWAN technology, incorporating a twisted Artificial Intelligence (AI) Model that integrates Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks intertwining their strengths to achieve automatic brightness control and power consumption optimization. The CNN in the model is responsible for accurately analysing current brightness and traffic flow levels, while the LSTM is used to predict future visibility changes. With the combination, the system can be flexible in dynamic environments and cope with changes in severe weather by adjusting the streetlights’ brightness in advance. This research utilizes the traffic road image dataset from the Kaggle platform and the weather dataset from the Szeged area to simulate the real environment. During the 100-day simulation, the average daily power consumption of ISLS was reduced by 13.98