IoT devices enable a massive amount of data to be aggregated and analyzed for anomaly detection. The nature of heterogeneous devices introduces the challenge of collecting and handling these massive datasets to perfor...
详细信息
Vehicular network services in the smart cities generate enormous data by vehicular road users, which is a critical challenge. Network traffic leads to a negative impact on safety applications. AI techniques are a prom...
Vehicular network services in the smart cities generate enormous data by vehicular road users, which is a critical challenge. Network traffic leads to a negative impact on safety applications. AI techniques are a promising solution to address network traffic in VANETs with V2X data. In this paper, we propose a soft voting classification model, which consists of hybrid supervised machine learning algorithms to predict traffic in the network. We evaluate the prediction performance of five well-known machine learning models and the proposed model based on various classification evaluation metrics. The simulation results show that the proposed network traffic prediction model performs better than other considered machine learning models in terms of accuracy (0.94%), time consumption (12.25 seconds) and AUROC (0.907) that proves its stability.
In the field of radiocommunication, modulation type identification is one of the most important characteristics in signal processing. This study aims to implement a modulation recognition system on two approaches to m...
详细信息
In the field of radiocommunication, modulation type identification is one of the most important characteristics in signal processing. This study aims to implement a modulation recognition system on two approaches to machine learning techniques, the K-Nearest Neighbors (KNN) and Artificial Neural Networks (ANN). From a statistical and spectral analysis of signals, nine key differentiation features are extracted and used as input vectors for each trained model. The feature extraction is performed by using the Hilbert transform, the forward and inverse Fourier transforms. The experiments with the AMC Master dataset classify ten (10) types of analog and digital modulations. AM_DSB_FC, AM_DSB_SC, AM_USB, AM_LSB, FM, MPSK, 2PSK, MASK, 2ASK, MQAM are put forward in this article. For the simulation of the chosen model, signals are polluted by the Additive White Gaussian Noise (AWGN). The simulation results show that the best identification rate is the MLP neuronal method with 90.5% of accuracy after 10 dB signal-to-noise ratio value, with a shift of more than 15% from the k-nearest neighbors’ algorithm.
Image segmentation is a fundamental component of either image processing or computer vision, finding its applications in medical image analysis, augmented reality, and video surveillance, among others. However, the cu...
详细信息
Image segmentation is a fundamental component of either image processing or computer vision, finding its applications in medical image analysis, augmented reality, and video surveillance, among others. However, the current research is paying too little attention to the robustness of such models, which is actually a factor that easily predisposes the model to adversarial perturbations caused by slight, imperceptible distortions added to the input images. In this work, we leverage Metamorphic Testing (MT) to evaluate and boost Segmentation models robustness. Our key innovation lies in using GA to intelligently evolve and optimize transformation sequences, systematically discovering the most effective combinations of spatial and spectral distortions while maintaining image fidelity. Our segmentation robustness metamorphic testing approach (SegRMT) generates adversarial examples that maintain the visual coherence of images while adhering to a predefined Peak Signal-to-Noise Ratio (PSNR) threshold, ensuring genuine disruptions. We use the Cityscapes dataset for our experiments, which consists of 5,000 images from diverse stereo video sequences in urban environments across 50 cities. Our findings show that by combining metamorphic testing and a genetic algorithm (GA), our approach can significantly reduce the mean Intersection over Union (mIoU) produced by the DeepLabV3 segmentation model to 6.4%, while other baseline adversaries decrease mIoU values between 21.7% and 8.5%. Other findings indicate that SegRMT and other baseline adversarial training achieve higher performance if training and testing occurred on their separate specific adversarial datasets, with mIoU values up to 73%. Other findings indicate that SegRMT adversarial training increases the mIoU of a segmentation model to 53.8% in cross-adversarial testings, while other baseline adversaries only increase mIoU values to between 2% and 10% on the SegRMT adversarial testing. This demonstrates that SegRMT effectiv
This paper introduces an Access Gate Function (AGF) that has been implemented using the P4 language and evaluates its performance when running on a Tofino switch. Through the process of translating specifications from...
This paper introduces an Access Gate Function (AGF) that has been implemented using the P4 language and evaluates its performance when running on a Tofino switch. Through the process of translating specifications from the Broadband Forum (BBF) for Fixed Mobile Convergence in the 5G Core into a P4 code and by conducting data plane simulations on software switch models, we provide evidence that the AGF solution conforms to industry standards and exhibits efficient and scalable performance characteristics. Subsequently, we assess the prototype’s implementation on a Tofino switch, analyzing its capabilities and limitations. The reported results demonstrate that the programmable P4 Tofino switch efficiently handles a significant number of sessions, further reinforcing the potential and practicality of the proposed solution for industrial deployment in fixed-mobile convergence scenarios.
Model stealing attacks have become a serious concern for deep learning models, where an attacker can steal a trained model by querying its black-box API. This can lead to intellectual property theft and other security...
详细信息
Model stealing attacks have become a serious concern for deep learning models, where an attacker can steal a trained model by querying its black-box API. This can lead to intellectual property theft and other security...
Model stealing attacks have become a serious concern for deep learning models, where an attacker can steal a trained model by querying its black-box API. This can lead to intellectual property theft and other security and privacy risks. The current state-of-the-art defenses against model stealing attacks suggest adding perturbations to the prediction probabilities. However, they suffer from heavy computations and make impracticable assumptions about the adversary. They often require the training of auxiliary models. This can be time-consuming and resource-intensive which hinders the deployment of these defenses in real-world applications. In this paper, we propose a simple yet effective and efficient defense alternative. We introduce a heuristic approach to perturb the output probabilities. The proposed defense can be easily integrated into models without additional training. We show that our defense is effective in defending against three state-of-the-art stealing attacks. We evaluate our approach on large and quantized (i.e., compressed) Convolutional Neural Networks (CNNs) trained on several vision datasets. Our technique outperforms the state-of-the-art defenses with a ×37 faster inference latency without requiring any additional model and with a low impact on the model's performance. We validate that our defense is also effective for quantized CNNs targeting edge devices.
Poorly managed postoperative acute pain can have long-lasting negative impacts and pose a major healthcare issue. There is limited investigation to understand and address the unique needs of patients experiencing acut...
Due to the risks associated with vulnerabilities in smart contracts, their security has gained significant attention in recent years. However, there is a lack of open datasets on smart contract vulnerabilities and the...
Due to the risks associated with vulnerabilities in smart contracts, their security has gained significant attention in recent years. However, there is a lack of open datasets on smart contract vulnerabilities and their fixes that allows for data-driven research. Towards this end, we propose an automated framework for mining and classifying Ethereum’s smart contract vulnerabilities and their corresponding fixes from GitHub and from the Common Vulnerabilities and Exposures (CVE) records in the National Vulnerability Database. We implemented the proposed method in a fully automated framework, which we call AutoMESC. AutoMESC uses seven of the most well-known smart contract security tools to classify and label the collected vulnerabilities based on vulnerability types. Furthermore, it collects metadata that can be used in data-intensive smart contract security research (e.g., vulnerability detection, vulnerability classification, severity prediction, and automated repair). We used AutoMESC to construct a sample dataset and made it publicly available. Currently, the dataset contains 6.7K smart contract vulnerability-fix pairs written in Solidity. We assess the quality of the constructed dataset in terms of accuracy, provenance, and relevance, and compare it with existing datasets. AutoMESC is designed to collect data continuously and keep the corresponding dataset up-to-date with newly discovered smart contract vulnerabilities and their fixes from GitHub and CVE records.
Context: Smart contracts are prone to numerous security threats due to undisclosed vulnerabilities and code weaknesses. In Ethereum smart contracts, the challenges of timely addressing these code weaknesses highlight ...
详细信息
暂无评论