Skin cancer is a serious health problem that, if not diagnosed correctly at an early stage, leads to disease progression, increased risk of metastasis, and decreased quality of life. Therefore, it is critical to devel...
详细信息
Critical embedded software, such as spacecraft flight software, requires high reliability, efficiency, and real-time performance while being capable enough to meet application needs. The increasing prevalence of Linux...
详细信息
A persistent issue of class imbalance in intrusion detection systems (IDS) is addressed in this research, particularly in the context of cyber-physical systems and IoT devices. Current IDS approaches often struggle to...
详细信息
ISBN:
(数字)9798350357509
ISBN:
(纸本)9798350357516
A persistent issue of class imbalance in intrusion detection systems (IDS) is addressed in this research, particularly in the context of cyber-physical systems and IoT devices. Current IDS approaches often struggle to detect rare attack patterns due to the dominance of normal traffic in imbalanced datasets. To address this, we present a unique strategy that combines Principal Component Analysis (PCA) for selecting features with the Synthetic Minority Oversampling Technique (SMOTE) to deal with class imbalance, applied to four powerful classifiers: Random Forest (RF), Decision Trees (DT), XGBoost (XGB), and AdaBoost. By reducing the feature space with PCA and balancing the dataset with SMOTE, our methodology significantly enhances model performance. Evaluating on the KDDCup ’99 dataset, we observe notable improvements: for instance, Random Forest achieves 96.42% accuracy, 98.52% precision, 94.68% recall 96.55% F1-score, and 93.87% MCC. Similarly, XGBoost achieves 94.78% accuracy, 96.72% precision, 92.94% recall, 94.81% F1-score, and 92.36% MCC. These results demonstrate that the integration of SMOTE and PCA effectively addresses class imbalance and enhances the detection of minority class attacks. Our findings contribute to the advancement of machine learning techniques in IDS and provide a foundation for further research into combining oversampling methods and dimensionality reduction for robust cybersecurity solutions. Future work will focus on exploring alternative oversampling methods, incorporating deep learning models, and assessing the practical deployment of the proposed methodology in real-time network environments.
Model interpretability has become increasingly critical in artificial intelligence research, particularly for high-stakes applications where transparency and trustworthiness are paramount. A fundamental challenge in t...
详细信息
Model interpretability has become increasingly critical in artificial intelligence research, particularly for high-stakes applications where transparency and trustworthiness are paramount. A fundamental challenge in this field emerges from an intriguing phenomenon: models that achieve comparable predictive performance often yield inconsistent feature importance scores (attribution scores interpretations) for identical data. This interpretability dilemma, manifest in both the Rashomon effect set (diverse models with different architectures achieving similar performance) and its subset, the Underspecification set (identical architectures varying due to training randomness), significantly diminishes the credibility of models' explanations. To address this challenge, this paper firstly explores theoretically the relationship between explanation inconsistency and model indeterminacy factors, and then proves that due to the local accuracy property of SHAP (SHapley Additive exPlanation), there exists an upper bound on the uncertainty of SHAP methods for models with similar predictions. On this basis, we thoroughly investigate experimentally the specific impacts of variables (i. e., model training stochastic factors) on various feature attribution methods. It finds that explanation uncertainty arising from model indeterminacy is widespread, whereas SHAP methods exhibit lower uncertainty due to the impact of its upper bound. Based on these findings, we propose an explanation rectification framework called ASGM (Attribution Score Generation Method) to generate stable attribution score explanations using standard explanations obtained from diverse models, aiming to reduce the inconsistency of attribution score explanations and enhance the stability and credibility of model interpretations. ASGM identifies disparities between explanations from several sampled models and standard explanations generated from massive models on specified data. Through bias rectification deep networ
Heavy metal wastewater contamination has become one of the greatest global environmental problems. In this study, magnetic Laponite/poly(AA-AM) composite hydrogels (mLap/P(AM–AA)) with multi-level three-dimensional n...
详细信息
Offshore structures are continuously subjected to wave-induced loads, resulting in both rigid-body motion and elastic deformation. In very large floating structures (VLFSs), hydroelastic deformation is primarily chara...
详细信息
The demand for precise timing continues to grow with the rapid development of smart cities, where intelligent transportation emerges as a critical application heavily dependent on accurate time distribution. Determini...
详细信息
ISBN:
(数字)9798350368369
ISBN:
(纸本)9798350368376
The demand for precise timing continues to grow with the rapid development of smart cities, where intelligent transportation emerges as a critical application heavily dependent on accurate time distribution. Deterministic communication is anticipated to be a defining feature of nextgeneration mobile networks (6G), enabling end-to-end timecritical applications, and unlocking new possibilities for achieving end-to-end precise time synchronization as well. To cope with the challenge of delivering precise time synchronization across broader areas at a lower cost, this paper proposes a method to achieve end-to-end precise time synchronization by leveraging the deterministic characteristics of communication networks. Specially, the proposed approach adopts the software-Defined Networking (SDN) paradigm to implement deterministic networking for the seamless integration of Time-Sensitive Networking (TSN) and cellular networks. Furthermore, to mitigate the impact of wireless channel uncertainties on synchronization accuracy, we propose a network delay measurement mechanism based on the principle of time redundancy, designed to reduce end-to-end network delay variation. Experimental results validate the effectiveness of the proposed method, demonstrating its capability to constrain the end-to-end relative time error of wired and wireless converged networks to the microsecond scale, without necessitating modifications to the existing network infrastructure.
This work presents the starfish optimization algorithm (SFOA), a novel bio-inspired metaheuristic for solving optimization problems, which simulates behaviors of starfish, including exploration, preying, and regenerat...
详细信息
The latest Segment Anything Model enables realtime scene annotation and understanding for autonomous driving systems, enhancing driving safety. However, effectively allocating resources for real-time performance and a...
详细信息
ISBN:
(数字)9798350368741
ISBN:
(纸本)9798350368758
The latest Segment Anything Model enables realtime scene annotation and understanding for autonomous driving systems, enhancing driving safety. However, effectively allocating resources for real-time performance and accuracy remains challenging in edge-cloud architectures. Traditional reinforcement learning struggles with poor generalization and the explorationexploitation dilemma, making it difficult to define clear reward functions. To address this, we propose a likelihood active inference approach to optimize resource allocation and improve system resource utilization. We use "intelligence" as a high-level indicator to quantify the efficiency of cognition in active inference, evaluating the difference between predicted and actual states during policy exploration. Experimental results show our algorithm outperforms mainstream deep reinforcement learning algorithms, improving sample efficiency and suitability for dynamically changing task workloads.
Ascribe to the research significance and great application potential of elastic wave manipulation, the topological phononic crystals with peculiar functions in robust waveguiding have attracted enormous research atten...
详细信息
暂无评论