PeerJ Computer Science最新文献

With the rapid development and widespread adoption of intelligent vehicles and the Internet of Vehicles (IoV), vehicle security has become a growing concern. Modern vehicles manage key components via the controller area network (CAN) connected electronic control units (ECUs). CAN bus intrusion techniques are the primary methods of compromising the IoV, posing a significant threat to the normal operation of critical vehicle systems, such as the power systems. However, existing attack detection methods still have shortcomings in terms of feature extraction and the diversity of attack type detection. To address these challenges, we propose an intrusion detection framework named basic ensemble and pioneer class decision (BEPCD). The framework first constructs a 15-dimensional feature model to hierarchically characterize CAN bus messages. Subsequently, BEPCD incorporates multi-model ensemble learning enhanced by a Pioneer class selector and confidence-driven voting mechanisms, enabling precise classification of both conventional and emerging attack patterns. Additionally, we analyze the importance of different data features across four machine learning algorithms. Experimental results on public datasets demonstrate that the proposed detection framework effectively detects intrusions in-vehicle CAN bus. Compared to other intrusion detection frameworks, our framework improves the overall F1-score by 1% to 5%. Notably, it achieves an approximately 77.5% performance enhancement in detecting replay attacks.

With the increasing prevalence and diversity of imaging devices, palmprint recognition has emerged as a technology that better meets the demands of the modern era. However, traditional manual methods have limitations in effectively extracting palmprint principal line features. To address this, we introduce a novel data augmentation method. First, the wide line extraction (WLE) filter is utilized to specifically target and extract the prominent principal lines of palmprints by leveraging their direction and width characteristics. Then, a Gabor filter is applied to the WLE-extracted results to purify the features and remove fine lines, as fine lines can introduce noise and redundancy that interfere with the accurate extraction of significant principal line features crucial for palmprint recognition. Evaluating this data augmentation across four common Vision Transformer (ViT) classification models, experimental results show that it improves the recognition rates of all databases to varying degrees, with a remarkable 32.9% increase on the high-resolution XINHUA database. With the successful removal of fine lines by WLE, we propose a new Layer Visual Transformer (LViT) design paradigm. For its input, distinct blocking strategies are adopted, carefully designed to partition the data to capture different levels of spatial and feature information, using larger blocks for global structure and smaller ones for local details. The output results of these different blocking strategies are fused by "sum fusion" and "maximum fusion", and the local and global features are effectively utilized by combining complementary information to improve the recognition performance and get state-of-the-art results on multiple databases. Moreover, LViT requires fewer training iterations due to the synergistic effects of the blocking strategies, optimizing the learning process. Finally, by simulating real-world noise conditions, we comprehensively evaluate LViT and find that, compared with traditional methods, our approach exhibits excellent noise-resistant generalization ability, maintaining stable performance across the PolyU II, IIT Delhi, XINHUA, and NTU-CP-V1 databases.

One of the main challenges encountered in visual object detection is the multi-scale issue. Many approaches have been proposed to tackle this issue. In this article, we propose a novel neck that can perform effective fusion of multi-scale features for a single-stage object detector. This neck, named the deformable convolution and path aggregation network (DePAN), is an integration of a path aggregation network with a deformable convolution block added to the feature fusion branch to improve the flexibility of feature point sampling. The deformable convolution block is implemented by repeated stacking of a deformable convolution cell. The DePAN neck can be plugged in and easily applied to various models for object detection. We apply the proposed neck to the baseline models of Yolov6-N and YOLOV6-T, and test the improved models on COCO2017 and PASCAL VOC2012 datasets, as well as a medical image dataset. The experimental results verify the effectiveness and applicability in real-world object detection.

The rapid advancement of artificial intelligence (AI) has catalyzed transformative changes in education, particularly in mobile and online learning environments. While existing deep learning models struggle to efficiently integrate the complexity of remote education data and optimize model performance, this article proposes an intelligent evaluation method for students' learning states based on multimodal data. First, the joint characteristics of the pre-class mental status survey information and the health big data of teachers and students in the online teaching process constitute input data. Then, the multilayer perceptron (MLP) is used to intelligently identify the students' status and classify their enthusiasm for the class. Finally, the particle swarm optimization (PSO) model is used to optimize the model and improve the overall recognition rate. Compared to traditional methods, the PSO-MLP model with combined multimodal data performs well, achieving an accuracy of 0.891. It provides an operational, technical solution for the education system, provides a new AI foundation for personalized teaching and student health management by accurately assessing students' learning status, and helps to improve the effectiveness and efficiency of remote education.

Microorganisms play an important role in many complex diseases, influencing their onset, progression, and potential treatment outcomes. Exploring the associations between microbes and human diseases can deepen our understanding of disease mechanisms and assist in improving diagnosis and therapy. However, traditional biological experiments used to uncover such relationships often demand substantial time and resources. In response to these limitations, computational methods have gained traction as more practical tools for predicting microbe-disease associations. Despite their growing use, many of these models still face challenges in terms of accuracy, stability, and adaptability to noisy or sparse data. To overcome the aforementioned limitations, we propose a novel predictive framework, HyperGraph Neural Network with Transformer for Microbe-Disease Associations (HGNNTMDA), designed to infer potential associations between human microbes and diseases. The framework begins by integrating microbe-disease association data with similarity-based features to construct node representations. Two graph construction strategies are employed: a K-nearest neighbor (KNN)-based adjacency matrix to build a standard graph, and a K-means clustering approach that groups similar nodes into clusters, which serve as hyperedges to define the incidence matrix of a hypergraph. Separate hypergraph neural networks (HGNNs) are then applied to microbe and disease graphs to extract structured node-level features. An attention mechanism (AM) is subsequently introduced to emphasize informative signals, followed by a Transformer module to capture contextual dependencies and enhance global feature representation. A fully connected layer then projects these features into a unified space, where association scores between microbes and diseases are computed. For model optimization, we propose a hybrid loss strategy combining contrastive loss and Huber loss. The contrastive loss aids in learning discriminative embeddings, while the Huber loss enhances robustness against outliers and improves predictive stability. The effectiveness of HGNNTMDA is validated on two benchmark datasets-HMDAD and Disbiome-using five-fold cross-validation (5CV). Our model achieves an AUC of 0.9976 on HMDAD and 0.9423 on Disbiome, outperforming six existing state-of-the-art methods. Further case studies confirm its practical value in discovering novel microbe-disease associations.

The rapid expansion of the Internet of Things (IoT) has significantly increased the volume and diversity of network traffic, making accurate IoT traffic classification crucial for maintaining network security and efficiency. However, existing traffic classification methods, including traditional machine learning and deep learning approaches, often exhibit critical limitations, such as insufficient generalization across diverse IoT environments, dependency on extensive labelled datasets, and susceptibility to overfitting in dynamic scenarios. While recent transformer-based models show promise in capturing contextual information, they typically rely on standard tokenization, which is ill-suited for the irregular nature of IoT traffic and often remains confined to single-purpose tasks. To address these challenges, this study introduces MIND-IoT, a novel and scalable framework for classifying generalized IoT traffic. MIND-IoT employs a hybrid architecture that combines Transformer-based models for capturing long-range dependencies and convolutional neural networks (CNNs) for efficient local feature extraction. A key innovation is IoT-Tokenize, a custom tokenization pipeline designed to preserve the structural semantics of network flows by converting statistical traffic features into semantically meaningful feature-value pairs. The framework operates in two phases: a pre-training phase utilizing masked language modeling (MLM) on large-scale IoT data (UNSW IoT Traces and MonIoTr) to learn robust representations and a fine-tuning phase that adapts the model to specific classification tasks, including binary IoT vs. non-IoT classification, IoT category classification, and device identification. Comprehensive evaluation across multiple diverse datasets (IoT Sentinel, YourThings, and IoT-FCSIT, in addition to the pre-training datasets) demonstrates MIND-IoT's superior performance, robustness, and adaptability compared to traditional methods. The model achieves an accuracy of up to 98.14% and a 97.85% F1-score, demonstrating its ability to classify new datasets and adapt to emerging tasks with minimal fine-tuning and remarkable efficiency. This research positions MIND-IoT as a highly effective and scalable solution for real-world IoT traffic classification challenges.

One of the most prominent neurodegenerative diseases globally is Alzheimer's disease (AD). The early diagnosis of AD is a challenging task due to complex pathophysiology caused by the presence and accumulation of neurofibrillary tangles and amyloid plaques. However, the late enriched understanding of the genetic underpinnings of AD has been made possible due to recent advancements in data mining analysis methods, machine learning, and microarray technologies. However, the "curse of dimensionality" caused by the high-dimensional microarray datasets impacts the accurate prediction of the disease due to issues of overfitting, bias, and high computational demands. To alleviate such an effect, this study proposes a gene selection approach based on the parameter-free and large-scale manta ray foraging optimization algorithm. Given the dimensional disparities and statistical relationship distributions of the six investigated datasets, in addition to four evaluated machine learning classifiers; the proposed Sign Random Mutation and Best Rank enhancements that substantially improved MRFO's exploration and exploitation contributed to efficient identification of relevant genes and to machine learning improved prediction accuracy.

As the core engine of electronic design automation (EDA) tools, the efficiency of Boolean Satisfiability Problem (SAT) solver largely determines the cycle of integrated circuit research and development. The effectiveness of SAT solvers has steadily turned into the key bottleneck of circuit design cycle due to the dramatically increased integrated circuit scale. The primary issue of SAT solver now is the divergence between SAT used in industry and research on pure solution algorithms. We propose a strategy for partitioning the SAT problem based on the structural information then solving it. By effectively extracting the structure information from the original SAT problem, the self-organizing map (SOM) neural network deployed in the division section can speed up the sub-thread solver's processing while avoiding cumbersome parameter adjustments. The experimental results demonstrate the stability and scalability of our technique, which can drastically shorten the time required to solve industrial benchmarks from various sources.

With the rapid development of the Internet of Things, time series classification (TSC) has gained significant attention from researchers due to its applications in various real-world fields, including electroencephalogram/electrocardiogram classification, emotion recognition, and error message detection. To improve classification performance, numerous TSC methods have been proposed in recent years. Among these, shapelet-based TSC methods are particularly notable for their intuitive interpretability. A critical task within these methods is evaluating the quality of candidate shapelets. This paper provides a comprehensive survey of the state-of-the-art measures for assessing shapelet quality. To present a structured overview, we begin by proposing a taxonomy of these measures, followed by a detailed description of each one. We then discuss these measures, highlighting the challenges faced by current research and offering suggestions for future directions. Finally, we summarize the findings of this survey. We hope that this work will serve as a valuable resource for researchers in the field.

Intensity modulated radiation therapy (IMRT) is a prevalent approach for administering radiation therapy in cancer treatment. The primary objective of IMRT is to devise a treatment strategy that eradicates cancer cells from the tumour while minimising damage to the surrounding organs at risk. Conventional IMRT planning entails a sequential procedure: optimising beam intensity for a certain set of angles, followed by sequencing. Unfortunately, treatment plans obtained in the optimisation stage are severely impaired after the sequencing stage due to physical and delivery constraints that are not considered during the optimisation stage. One method that tackles the issues above is the direct aperture optimisation (DAO) technique. The DAO problem seeks to generate a set of deliverable aperture configurations and a corresponding set of radiation intensities. This method accounts for physical and delivery time limitations, facilitating the creation of clinically appropriate treatment programs. In this article, we propose and compare two variable neighbourhood search (VNS) based algorithms, called variable neighbourhood descent (VND) and reduced variable neighbourhood search (rVNS). The VND algorithm is a deterministic variant of VNS that systematically explores different neighbourhood structures. This approach allows for a more thorough solution for space exploration while maintaining computational efficiency. The rVNS, unlike traditional VNS algorithms, does not require any transition rule, as it integrates a set of predefined neighbourhood moves at each iteration. We apply our proposed algorithms to prostate cancer cases, achieving highly competitive results for both algorithms. In particular, the proposed rVNS requires 62.75% fewer apertures and achieved a 63.93% reduction in beam-on time compared to the sequential approach's best case, which means treatment plans that can be delivered in less time. Additionally, we evaluate the clinical quality of the treatment plans using established dosimetric indicators, comparing our results against those produced by matRad's tool for DAO to assess target coverage and organ-at-risk sparing.