Neural Computing and Applications最新文献

Portable document formats (PDFs) are widely used for document exchange due to their widespread usage and versatility. However, PDFs are highly vulnerable to malware attacks, which pose significant security risks. Existing defense mechanisms often struggle to effectively detect and mitigate these threats, highlighting the need for more robust solutions. This paper introduces a robust framework that uses advanced tree-based ensemble models to detect malicious PDFs using the Evasive-PDFMal2022 dataset. The proposed model achieves a recall rate of 100%, an accuracy rate of 99.95%, and a fast inference time of 0.1723 s. Furthermore, the framework exhibits minimal false positive and false negative rates, ensuring a high level of precision in distinguishing between malicious and benign PDFs. Shapley additive explanations are used to improve the interpretability and reliability of the model’s predictions. The results highlight the effectiveness of the proposed model in improving PDF document security and addressing the challenges posed by malware attacks.

In anomaly detection applications, anomalies might come from multiple sources and there might be many reasons why a sample is considered to be anomalous. However, most novel anomaly detection methods do not consider this. In our work, we describe a novel approach that is demonstrated on the problem of detection of anomalies in image data. We propose the SGVAEGAN model, which decomposes the image into three independent components—the shape of an object and its foreground and background textures—and provides anomaly scores for each of those factors separately. The overall anomaly score of an image is a weighted combination of the individual factor scores. The anomaly scores are learned in an unsupervised manner, and the weights are considered as hyperparameters that can be learned in the validation stage. The approach allows the identification of the source of the anomaly using factor scores, as well as the detection of semantic anomalies where the semantic meaning is encoded in the weights and learned from very few samples of validation anomalies. On classical anomaly detection benchmarks, the proposed model outperforms all baseline models. This is shown in a rigorous experimental study that covers the behavior of the model under a varying range of conditions.

This study proposes a novel multimodal approach for mixed-frequency time series forecasting in the oil industry, enabling the use of high-frequency (HF) data in their original frequency. We specifically address the challenge of integrating HF data streams, such as pressure and temperature measurements, with daily time series without introducing noise. Our approach was compared with existing econometric regression model mixed-data sampling (MIDAS) and with the data-driven models N-HiTS and a GRU-based network, across short-, medium-, and long-term prediction horizons. Additionally, we validated the proposed method on datasets from other domains beyond the oil industry. The experimental results indicate that our multimodal approach significantly improves long-term prediction accuracy.

Decision-making is crucial in fully autonomous vehicle operations and is expected to greatly influence future transportation systems. Observing the current driving status of autonomous vehicles is vital for its decision-making process. The autonomous connected vehicles on the road send significant data about their movements to the server to maintain continuous training. With the Proof of Authority (PoA) consensus process, blockchain technology provides a valid, decentralised and secure option to improve transactions throughput and minimise delay. The limited computational capacity of vehicles poses a challenge in achieving high accuracy and low latency while training self-driving algorithms. GPT-4V surpassed challenging autonomous systems in scene interpretation and causal thinking. GPT-4V has ability to navigate circumstances without access to database, interpret intentions, and make sound decisions in real-world driving scenarios. The reward function and different driving conditions are organised to allow an optimal search to find the most efficient driving style while ensuring safety. The consequences of the Blockchain-enabled decision-making model (DMM) for Self-Driving Vehicles (SDV) primarily based on GPT-4V and Federated Reinforcement Learning (FRL) would, likely, upgrades in decision-making accuracy, operational performance, statistics integrity, and potentially enhanced learning skills in SDV. Integrating blockchain technology, superior language modelling GPT-4V and FRL may lead to multiplied safety, reliability, and decision-making ability in SDV. This study utilised the Simulation of Urban MObility (SUMO) simulator to assess the ability of SDV to maintain its desired speed consistently and securely in a highway setting using proposed DMM. This study indicates that the suggested DMM, utilising the driving state evaluation approach for SDV, can help these vehicles operate safely and effectively. The performance of the proposed model, such as CPU utilisation, bandwidth and latency, are evaluated through multiple tests.

Fifth generation (5G) mobile networks enable ultra-reliable low-latency communication (URLLC) applications, ushering in an era of endless possibilities for 5G. URLLC supports emerging 5G services and applications with stringent requirements for latency and reliability. Factory automation (FA) is a URLLC application that automates and optimizes workflows and processes in factories. To accommodate diversified FA services, 5G networks employ the “network slicing” technique, which divides the network into slices tailored to different service requirements. Designing a sliced network and translating diversified service-level agreements (SLAs) into network attributes necessitates advanced automation techniques to enhance human–machine collaboration, increase efficiency, minimize manual errors, reduce operating costs, and, most importantly, provide adequate service quality economically and reliably. To apply autonomic computing to FA network design, new architectures and software components have been envisioned. These include information extraction, domain knowledge representation, rule-based reasoning, performance model calculation, and querying using simulators and neural networks (NNs), among others. This paper proposes an innovative approach to network slicing design using advanced automation methods. This approach can be easily extended to include new services or to integrate cutting-edge 5G techniques.

Recent advancements in deep neural networks (DNNs) have made them indispensable for numerous commercial applications. These include healthcare systems and self-driving cars. Training DNN models typically demands substantial time, vast datasets and high computational costs. However, these valuable models face significant risks. Attackers can steal and sell pre-trained DNN models for profit. Unauthorised sharing of these models poses a serious threat. Once sold, they can be easily copied and redistributed. Therefore, a well-built pre-trained DNN model is a valuable asset that requires protection. This paper introduces a robust hybrid two-level protection system for safeguarding the ownership of pre-trained DNN models. The first-level employs zero-bit watermarking. The second-level incorporates an adversarial attack as a watermark by using a perturbation technique to embed the watermark. The robustness of the proposed system is evaluated against seven types of attacks. These are Fast Gradient Method Attack, Auto Projected Gradient Descent Attack, Auto Conjugate Gradient Attack, Basic Iterative Method Attack, Momentum Iterative Method Attack, Square Attack and Auto Attack. The proposed two-level protection system withstands all seven attack types. It maintains accuracy and surpasses current state-of-the-art methods.

In the recent years, neuroimaging and deep learning have received notable scientific attention for the diagnosis of grade IV tumor de novo glioblastoma in the central nervous system. However, the scarce amount of neuroimaging data for training has resulted in significant overfitting issues for numerous deep learning models. To address these challenges, we propose the implementation of a meta-learning-based IRAM–NET model that utilizes the ResNet-50 as a deep learning-based model and incorporates the e-MAML ensemble technique from meta-learning for the early diagnosis of glioblastoma. The methodology developed was trained and validated using brain MRI images taken from numerous national and international cancer initiative data repositories. In the training phase, this study employed detailed procedures, including the handling of exceptions and the application of normalization techniques. These measures were implemented to guarantee precise data representation, mitigate the risk of overfitting, and enhance the proposed model’s capacity for making meaningful generalizations. The proposed IRAM–NET model surpasses the most recent studies in accurately predicting glioblastoma diagnosis, achieving a training, testing and validation accuracy of 97.22%, 96.10%, and 94.74%, respectively. Overall, the research not only enhances the diagnosis of rare disorders like glioblastoma, but also promotes the wider inclusion of meta-learning in healthcare. This underlines the importance of adaptation and efficiency in situations with limited data availability.

Individuals who are younger and have dementia often start experiencing its symptoms before they turn 65, with cases even documented in people as young as their thirties. Researchers strive for accurate dementia diagnosis to slow or halt its progression. This paper presents a novel Enhanced Dementia Detection and Classification Model (EDCM) comprised of four modules: data acquisition, preprocessing, hyperparameter optimization, and feature extraction/classification. Notably, the model uses texture information from segmented brain images for improved feature extraction, leading to significant gains in both binary and multi-class classification. This is achieved by selecting optimal features via a Gray Wolf Optimization (GWO)-driven enhancement model. Results demonstrate substantial accuracy improvements after optimization. For instance, using an Extra Tree Classifier for "normal" cases, the model achieves 85% accuracy before optimization. However, with GWO-optimized features and hyperparameters, the accuracy jumps to 97%.

With the advancement in web interactions and increased use of Online Social Networks, sentiment analysis has gained popularity. Topics like sports, health, music, and technology are widely debated on in OSN, especially on twitter. People share their activities, views, and feelings toward different events in their native languages that can be analyzed using sentiment analysis to understand the sentiments of the people toward these events. For English language, studies on sentiment analysis are vastly available. However, very little work exists on sentiment analysis for resource-scarce language like Urdu. For this study, we perform aspect-based sentiment analysis on sports tweets in Urdu language by extracting the following information from a sentence, i.e., aspect terms, aspect term polarity, aspect category, and aspect category polarity, using machine learning and deep learning classifiers. This work is the first effort in aspect-based sentiment analysis for Urdu language using classical machine learning and deep learning approach. Additionally, we also identify implicit aspects from a sentence. Our proposed approach shows classical machine learning approach performed better on the tasks of aspect term polarity, aspect category, and aspect category polarity, while deep learning model outperformed classical machine learning classifiers for the task of aspect term/s.

The widespread use of online social networks (OSNs) has made them prime targets for cyber attackers, who exploit these platforms for various malicious activities. As a result, a whole industry of black-market services has emerged, selling services based on the sale of fake accounts. Because of the massive rise of OSNs, the number of fraudulent accounts rapidly expands. Hence, this research focuses on detecting fraudulent profiles on Instagram and Facebook and aims to find an optimal subset of features that can effectively differentiate between real and fake accounts. The problem has been formulated as a multiobjective optimization task, aiming to maximize the classification accuracy while minimizing the number of selected features. NSGA-II (non-dominated sorting genetic algorithm II) is employed as the optimization algorithm to explore the trade-offs between these conflicting objectives. In the current study, a novel approach for feature selection using the NSGA-II optimization algorithm to detect fake accounts is proposed. The proposed methodology relies on input data comprising features characterizing the profiles under investigation. The selected features are utilized to train a machine learning model. The model’s performance is evaluated using various metrics, including precision, recall, F1-score, and receiver operating characteristic (ROC) curve. The final prediction model achieved accuracy values ranging from 90 to 99.88%. The results indicated that the model, utilizing features selected by the NSGA-II algorithm, delivered high prediction accuracy while using less than 31% of the total feature space. This efficient feature selection allowed for the precise differentiation between fake and real users, demonstrating the model’s effectiveness with a minimal number of input variables. Furthermore, the results of experiments demonstrate that the proposed approach achieves better performance as compared to other existing approaches. This research paper focuses on explainability, which refers to the ability to understand and interpret the decisions and outcomes of machine learning models.