International Journal of Intelligent Systems最新文献

Eye tracking has emerged as a valuable tool for both research and clinical applications. However, traditional eye-tracking systems are often bulky and expensive, limiting their widespread adoption in various fields. Smartphone eye tracking has become feasible with advanced deep learning and edge computing technologies. However, the field still faces practical challenges related to large-scale datasets, model inference speed, and gaze estimation accuracy. The present study created a new dataset that contains over 3.2 million face images collected with recent phone models and presents a comprehensive smartphone eye-tracking pipeline comprising a deep neural network framework (MGazeNet), a personalized model calibration method, and a heuristic gaze signal filter. The MGazeNet model introduced a linear adaptive batch normalization module to efficiently combine eye and face features, achieving the state-of-the-art gaze estimation accuracy of 1.59 cm on the GazeCapture dataset and 1.48 cm on our custom dataset. In addition, an algorithm that utilizes multiverse optimization to optimize the hyperparameters of support vector regression (MVO–SVR) was proposed to improve eye-tracking calibration accuracy with 13 or fewer ground-truth gaze points, further improving gaze estimation accuracy to 0.89 cm. This integrated approach allows for eye tracking with accuracy comparable to that of research-grade eye trackers, offering new application possibilities for smartphone eye tracking.

Collaborative edge and cloud computing is a promising computing paradigm for reducing the task response delay and energy consumption of devices. In this paper, we aim to jointly optimize task offloading strategy, power control for devices, and resource allocation for edge servers within a collaborative device-edge-cloud computing system. We formulate this problem as a constrained multiobjective optimization problem and propose a joint optimization algorithm (JO-DEC) based on a multiobjective evolutionary algorithm to solve it. To address the tight coupling of the variables and the high-dimensional decision space, we propose a decoupling encoding strategy (DES) and a boundary point sampling strategy (BPS) to improve the performance of the algorithm. The DES is utilized to decouple the correlations among decision variables, and BPS is employed to enhance the convergence speed and population diversity of the algorithm. Simulation results demonstrate that JO-DEC outperforms three state-of-the-art algorithms in terms of convergence and diversity, enabling it to achieve a smaller task response delay and lower energy consumption.

Application programming interface (API) misuse refers to misconceptions or carelessness in the anticipated usage of APIs, threatening the software system’s security. Moreover, API misuses demonstrate significant concealment and are challenging to uncover. Recent advancements have explored enhanced LLMs in a variety of software engineering (SE) activities, such as code repair. Nonetheless, the security implications of using LLMs for these purposes remain underexplored, particularly concerning the issue of API misuse. In this paper, we present an empirical study to observe the bug-fixing capabilities of LLMs in addressing API misuse related to monitoring resource management (MRM API misuse). Initially, we propose APImisRepair, a real-world benchmark for repairing MRM API misuse, including buggy programs, corresponding fixed programs, and descriptions of API misuse. Subsequently, we assess the performance of several LLMs using the APImisRepair benchmark. Findings reveal the vulnerabilities of LLMs in repairing MRM API misuse and find several reasons, encompassing factors such as fault localization and a lack of awareness regarding API misuse. Additionally, we have insights on improving LLMs in terms of their ability to fix MRM API misuse and introduce a crafted approach, APImisAP. Experimental results demonstrate that APImisAP exhibits a certain degree of improvement in the security of LLMs.

Federated learning (FL) is a machine learning technique in which a large number of clients collaborate to train models without sharing private data. However, FL’s integrity is vulnerable to unreliable models; for instance, data poisoning attacks can compromise the system. In addition, system preferences and resource disparities preclude fair participation by reliable clients. To address this challenge, we propose a novel client selection strategy that introduces a security-fairness value to measure client performance in FL. The value in question is a composite metric that combines a security score and a fairness score. The former is dynamically calculated from a beta distribution reflecting past performance, while the latter considers the client’s participation frequency in the aggregation process. The weighting strategy based on the deep deterministic policy gradient (DDPG) determines these scores. Experimental results confirm that our method fairly effectively selects reliable clients and maintains the security and fairness of the FL system.

The conventional K-Means clustering algorithm is widely used for grouping similar data points by initially selecting random centroids. However, the accuracy of clustering results is significantly influenced by the initial centroid selection. Despite different approaches, including various K-Means versions, suboptimal outcomes persist due to inadequate initial centroid choices and reliance on common normalization techniques like min-max normalization. In this study, we propose an improved algorithm that selects initial centroids more effectively by utilizing a novel formula to differentiate between instance attributes, creating a single weight for differentiation. We introduce a preprocessing phase for dataset normalization without forcing values into a specific range, yielding significantly improved results compared to unnormalized datasets and those normalized using min-max techniques. For our experiments, we used five real datasets and five simulated datasets. The proposed algorithm is evaluated using various metrics and an external benchmark measure, such as the Adjusted Rand Index (ARI), and compared with the traditional K-Means algorithm and 11 other modified K-Means algorithms. Experimental evaluations on these datasets demonstrate the superiority of our proposed methodologies, achieving an impressive average accuracy rate of up to 95.47% and an average ARI score of 0.95. Additionally, the number of iterations required is reduced compared to the conventional K-Means algorithm. By introducing innovative techniques, this research provides significant contributions to the field of data clustering, particularly in addressing modern data-driven clustering challenges.

Age-related macular degeneration (AMD) diagnosis using fundus images is one of the critical missions of the eye-care screening program in many countries. Various proposed deep learning models have been studied for this research interest, which aim to achieve the mission and outperform human-based approaches. However, research efforts are still required for the improvement of model classification accuracy, sensitivity, and specificity values. In this study, we proposed the model named as ViT-AMD, which is based on the latest Vision Transformer (ViT) structure, to diagnosis a fundus image as normal, dry AMD, or wet AMD types. Unlike convolution neural network models, ViT consists of the attention map layers, which show more effective performance for image classification task. Our training process is based on the 5-fold cross-validation and transfer learning techniques using Chula-AMD dataset at the Department of Ophthalmology, the King Chulalongkorn Memorial Hospital, Bangkok. Furthermore, we also test the performance of trained model using an independent image datasets. The results showed that for the 3-classes AMD classification (normal vs. dry AMD vs. wet AMD) on the Chula-AMD dataset, the averaged accuracy, precision, sensitivity, and specificity of our trained model are about 93.40%, 92.15%, 91.27%, and 96.57%, respectively. For result testing on independent datasets, the averaged accuracy, precision, sensitivity, and specificity of trained model are about 74, 20%, 75.35%, 74.13%, and 87.07%, respectively. Compared with the results from the baseline CNN-based model (DenseNet201), the trained ViT-AMD model has outperformed significantly. In conclusion, the ViT-AMD model have proved their usefulness to assist the ophthalmologist to diagnosis the AMD disease.

The networked control systems (NCSs) under cyberattacks have received much attention in both industrial and academic fields, with rare attention on the delayed networked control systems (DNCSs). In order to well address the control problem of DNCSs, in this study, we consider the resilient event-triggered safety control problem of the NCSs with time-varying delays based on the switched observer subject to aperiodic denial-of-service (DoS) attacks. The observer-based switched event-triggered control (ETC) strategy is devised to cope with the DNCSs under aperiodic cyberattacks for the first time so as to decrease the transmission of control input under limited network channel resources. A new piecewise Lyapunov functional is proposed to analyze and synthesize the DNCSs with exponential stability. The quantitative relationship among the attack activated/sleeping period, exponential decay rate, event-triggered parameters, sampling period, and maximum time-delay are explored. Finally, we use both a numerical example and a practical example of offshore platform to show the effectiveness of our results.

In most peer-to-peer (P2P) networks, peers are placed randomly or based on their geographical position, which can lead to a performance bottleneck. This problem can be solved by using peer clustering algorithms. In this paper, the significant results of the paper can be described in the following sentences. We propose two innovative swarm-based metaheuristics for peer clustering, slime mold and slime mold K-means. They are competitively benchmarked, evaluated, and compared to nine well-known conventional and swarm-based algorithms: artificial bee colony (ABC), ABC combined with K-means, ant-based clustering, ant K-means, fuzzy C-means, genetic K-means, hierarchical clustering, K-means, and particle swarm optimization (PSO). The benchmarks cover parameter sensitivity analysis and comparative analysis made by using 5 different metrics: execution time, Davies–Bouldin index (DBI), Dunn index (DI), silhouette coefficient (SC), and averaged dissimilarity coefficient (ADC). Furthermore, a statistical analysis is performed in order to validate the obtained results. Slime mold and slime mold K-means outperform all other swarm-inspired algorithms in terms of execution time and quality of the clustering solution.

Detecting deepfake media remains an ongoing challenge, particularly as forgery techniques rapidly evolve and become increasingly diverse. Existing face forgery detection models typically attempt to discriminate fake images by identifying either spatial artifacts (e.g., generative distortions and blending inconsistencies) or predominantly frequency-based artifacts (e.g., GAN fingerprints). However, a singular focus on a single type of forgery cue can lead to limited model performance. In this work, we propose a novel cross-domain approach that leverages a combination of both spatial and frequency-aware cues to enhance deepfake detection. First, we extract wavelet features using wavelet transformation and residual features using a specialized frequency domain filter. These complementary feature representations are then concatenated to obtain a composite frequency domain feature set. Furthermore, we introduce an adaptive feature fusion module that integrates the RGB color features of the image with the composite frequency domain features, resulting in a rich, multifaceted set of classification features. Extensive experiments conducted on benchmark deepfake detection datasets demonstrate the effectiveness of our method. Notably, the accuracy of our method on the challenging FF++ dataset is mostly above 98%, showcasing its strong performance in reliably identifying deepfake images across diverse forgery techniques.

Users’ online activities serve as a mirror, reflecting their unique personas, affiliations, interests, and hobbies within the real world. Network information dissemination is inherently targeted, as users actively seek information to facilitate precise and swift communication. Delving into the nuances of information propagation on the Internet holds immense potential for facilitating commercial endeavors such as targeted advertising, personalized product recommendations, and insightful consumer behavior analyses. Recognizing that the intensity of information transmission diminishes with the proliferation of competing messages, increased transmission distances, and the passage of time, this paper draws inspiration from the concept of heat attenuation to formulate an innovative information propagation model. This model simulates the “heat index” of each node in the transmission process, thereby capturing the dynamic nature of information flow. Extensive experiments, bolstered by comparative analyses of multiple datasets and relevant algorithms, validate the correctness, feasibility, and efficiency of our proposed algorithm. Notably, our approach demonstrates remarkable accuracy and stability, underscoring its potential for real-world applications.