In text classification tasks, deep neural networks are particularly susceptible to word-level adversarial attacks, where even small lexical perturbations can significantly change the predictions of the model. To address this challenge, this work proposes a novel defense framework, Gradient-based Synonym Substitution Attack - Adversarial Training (GSSA-AT), which integrates the Gradient-based Synonym Substitution Attack (GSSA) with adversarial training. Unlike other query-based methods, GSSA generates semantically consistent adversarial examples through a gradient-guided synonym substitution mechanism, while GSSA-AT leverages these examples alongside a loss function formulated in this paper to train the model and enhance its robustness. Experimental results demonstrate that GSSA-AT exhibits effective defense capabilities against various adversarial attacks, while also showing unique advantages in reducing the transferability of adversarial examples. Evaluations on three benchmark datasets reveal that GSSA-AT strikes an optimal balance between robustness and natural language understanding, ensuring the generalization of the model to clean data. These results establish GSSA-AT as an efficient and deployable defense solution for text classification.

Metadata, as information describing data, encompasses an exhaustive description of various facets of the data. Despite the enhanced accuracy of existing respiratory sound detection methods through multifaceted research, these methods often under-utilise metadata. To explore the potential of metadata, we study its impact on detection performance. We adopt a multi-supervised contrastive learning approach and propose an improved Metadata-Convolutional Neural Network model for more effective extraction of metadata features. We use the International Conference in Biomedical Health Informatics (ICBHI) 2017 database for evaluation and achieve an average score of 59.48

With the development of IoT and mobile sensing devices, trajectory data has an increasingly high research value. However, unauthorized data mining and analyzing will result in privacy violations. Therefore, the key issue is how to maintain user privacy while publishing usable data. To address the above problem, we propose a spatiotemporal trajectory data generalization publishing scheme with differential privacy (STG-DP), which consists of two components: trajectory processing and trajectory publishing. In trajectory processing, to improve data utility, a density-based trajectory clustering framework (DTC) is proposed, integrating two clustering algorithms to compare the impact of synthetic and real cluster centers on experimental results. In terms of trajectory publishing, an adaptive noise perturbation mechanism based on the staircase mechanism is proposed to enhance the degree of privacy protection. We theoretically prove that STG-DP satisfies the definition of differential privacy and experimentally verify it on a real dataset. The experiments demonstrate that STG-DP provides greater data utility and privacy protection than existing studies.

With the continuous escalation of cyberattacks, the forms of malware have become increasingly diverse, posing significant security threats to enterprises, government agencies, and individual users. Malware developers often employ techniques such as packing and obfuscation to evade detection, making traditional detection methods less effective. This study proposes a malware family classification method based on feature fusion and a two-layer classification framework. First, readable characters, bytes, and opcodes are extracted from the malware binary and disassembly files. Frequency and semantic features are extracted from both opcodes and readable characters, followed by frequency fusion and semantic fusion. A Markov image is generated from the byte transfer probability matrix. Next, a two-layer classification framework combining deep learning and traditional machine learning is designed based on the fused features and Markov image. This framework effectively integrates the advantages of different feature dimensions and models. In the first layer, each feature is detected, and in the second layer, the prediction probabilities of each feature are fused. Experimental results show that the proposed method achieves a malware family classification accuracy of 99.4

As the cybersecurity landscape continues to evolve, intrusion detection systems (IDS), a critical component of defense frameworks, face unprecedented challenges. One of the primary factors contributing to the decline in intrusion detection performance is the issue of class imbalance within datasets. To address this challenge, this paper proposes an intrusion detection method designed specifically for the class imbalance problem, named BOA-ACRF. This method first improves the traditional Auxiliary Classifier Generative Adversarial Network (ACGAN) to enhance its capability in generating numerical data for specific traffic categories. Furthermore, the Bayesian Optimization Algorithm (BOA) is employed to automatically identify optimal model parameters. This approach not only effectively resolves the sensitivity of ACGAN to hyperparameters but also improves the generalization capability and detection performance of the Random Forest (RF) model. The effectiveness of BOA-ACRF is validated on three intrusion detection datasets: CIC-IDS-2017, CIC-UNSW-NB15 and NSL-KDD. Experimental results show that the proposed method achieves outstanding performance in accuracy, precision, recall, and F1-score, significantly surpassing current mainstream approaches. This work provides an effective framework and technical solution to address the class imbalance problem in the field of intrusion detection.

Promoted by the Internet of Thing era, the widespread use of mobile sensing devices equipped with positioning functions has led to the generation of substantial trajectory data. Mining and analyzing trajectory data has high research value, but poses a risk of user privacy leakage, resulting in fewer publicly available trajectory datasets for research and analysis. Therefore, a trajectory data publishing method that ensures high data utility while protecting user privacy has become a hot topic. In this paper, we propose a spatiotemporal trajectory data protection and publishing method based on kernel density estimation (STP-KDE), which protects the trajectory data and trajectory count values while improving data utility. In the protection process of trajectory data, we designed a kernel density clustering framework that is combined with the differential privacy exponential mechanism. In the protection process of trajectory counts value, the adaptive Laplace noise perturbation mechanism is proposed to differentially protect these counts. Experimental results show that STP-KDE can provide more useful data and stronger privacy protection than existing studies.

In recent years, as cyberattacks continue to escalate, malware has become increasingly diverse and complex, posing significant security threats to enterprises, government agencies, and individual users. Malware developers often employ techniques such as feature obfuscation and behavior hiding, rendering traditional detection methods less effective. To address this challenge, this study proposes a malware detection method based on feature fusion and a multi-feature detection framework. The method extracts frequency features and semantic information from opcodes and readable characters, and byte transition probabilities from byte sequences, thereby constructing a comprehensive feature vector. A two-layer detection framework that combines deep learning with traditional machine learning is designed, effectively integrating different feature types and overcoming the limitations of single-feature approaches. Experimental results demonstrate that the proposed method significantly outperforms traditional algorithms in terms of detection accuracy and generalization capability, greatly enhancing the detection of complex malware families. Notably, it excels in handling packed code, obfuscation techniques, and imbalanced data, offering an efficient solution for malware detection.

With the rapid development of IoT technology, a large of number complex and diverse IoT devices are widely deployed, which brings new challenges for device identification due to the heterogeneous nature of devices. This paper proposes a network traffic-based IoT device recognition method, in order to solve the high cost problem of traditional recognition methods in the feature extraction process and the potential privacy leakage problem. The proposed method requires only a short period of time in the network traffic data of IoT devices, and extracts the protocol statistical features and flow-level statistical features of this data. It avoids in-depth inspection of packet payloads and reduces the cost of feature extraction effectively. It is demonstrated that the proposed method can improve the performance of device identification while ensuring privacy security through empirical studies on two widely recognized public datasets. The proposed method provides users with a low-cost, high-efficiency IoT device identification solution with strong privacy protection, which promotes wider and more secure application of IoT technology.

Android malware uses anti-reverse analysis and APK shelling technology, which leads to the failure of the classification method based on decompiled features and the reduction of the classification accuracy based on single file features. Moreover, the lack of samples in some families of Android malware makes the classification model based on sample learning ineffective. To solve the above problems, this paper proposes a two-layer general framework for Android malware classification and adversarial training named GMADV, which enhances classifier performance through adversarial training. In the sample classification layer, based on the transformation method of the Markov model, it is proposed for the first time to convert the three files in the APK into RGB Markov images, and use VGG13 to automatically extract features and classification; In the variant amplification layer, the idea of "regression for generation" is firstly proposed, and GMM-GAN based on Gaussian process is designed to amplify the diversity of samples within the family. The experimental results show that RGB Markov images have better classification performance than grayscale images. On the three datasets, the classification effect after amplification has been improved to varying degrees, and all F1_Score reaches 95 %. Compared with other methods, GMADV has stronger family sample amplification ability and greater adversarial intensity.

The domain name system is an essential part of the network, and target hosts are often attacked by malicious domain names to steal resources. Some traditional detection methods have low accuracy, poor generalization ability, and high resource overhead on model construction to deal with complex and variable malicious domain names. A three-level dynamic malicious domain detection method (DDOFM) is proposed in this paper. DDOFM only needs to combine a few high-order statistical features of benign domains with some DNS features, without flagging malicious samples and involving them in training. Firstly, the boundary recognition of passive DNS (PDNS) features extracted from DNS traffic is carried out to conduct an early- warning for some domains. Second, the Hidden Markov Model (HMM) forward algorithm and normal distribution probability density function are used to calculate the formation probabilities of the warned domains and their probability density values. Then the probabilities of every character in the warned domain name and the standard deviation between these probabilities are counted. Further, the probability density values and the standard deviations of these warned domain names are compared with their respective thresholds to identify the attribution of the warned domain names. Finally, if the domain name is not warned, the Jensen–Shannon divergence (JS divergence) between it and the previous domain name will be calculated. Then the local iterative threshold finding algorithm (LLTFA) proposed in this paper will be combined to determine its attribution and identify whether the host is connected to the command and control (C&C) server. Experiments show that the detection indexes of this method exceed 99% for multiple types of malicious domain names. The C&C servers can also be identified by DDOFM faster than similar methods.

Illegal elements use the characteristics of an anonymous network hidden service mechanism to build a dark network and conduct various illegal activities, which brings a serious challenge to network security. The existing anonymous traffic classification methods suffer from cumbersome feature selection and difficult feature information extraction, resulting in low accuracy of classification. To solve this problem, a classification method based on three-dimensional Markov images and output self-attention convolutional neural network is proposed. This method first divides and cleans anonymous traffic data packets according to sessions, then converts the cleaned traffic data into three-dimensional Markov images according to the transition probability matrix of bytes, and finally inputs the images to the output self-attention convolution neural network to train the model and perform classification. The experimental results show that the classification accuracy and F1-score of the proposed method for Tor, I2P, Freenet, and ZeroNet can exceed 98.5%, and the average classification accuracy and F1-score for 8 kinds of user behaviors of each type of anonymous traffic can reach 93.7%. The proposed method significantly improves the classification effect of anonymous traffic compared with the existing methods.

Malware is often embedded with domain generation algorithms (DGAs) to prevent firewall interception and domain black-and-white list comparison detection while hiding command and control (C&C) servers to tighten the control of botnets. DGA domains are diverse and difficult to obtain, resulting in highly unbalanced datasets. Domain names generated by different DGA families do not differ much at the sequence data level and it is difficult to extract their features. The above characteristics lead to poor accuracy, poor generalization ability, and bloatedness of DGA domain name classification models based on deep learning. To solve the above problems, the visual representation of sequence data and the DGA domain classification model are presented in this paper. First, the DGA domain name is mapped to the attention recurrence plot (Att_RP) proposed in this paper, which can enrich the data phase space features and differentiate the key phase space features. After that, Att_RP is sent to a DGA domain name classification model (CI_GRU) proposed in this paper for data dimension transformation processing, followed by classification. Experiments show that the classification accuracy, F1_score, and recall of the model for a variety of DGA families in the wild are higher than 99%, and can also accurately classify four types of crafted DGA families. Compared with similar models, the model has high classification accuracy, low time consumption, low generalization error, and high efficiency, and the size of the model is less than one-tenth of similar models.

为解决现有恶意代码检测方法存在的特征提取能力不足、检测模型泛化性弱的问题,提出了一种基于Windows API调用序列的恶意代码检测方法.使用N-gram算法和TF-IDF算法提取序列的统计特征,采用Word2Vec模型提取语义特征,将统计特征和语义特征进行特征融合,作为API调用序列的特征.设计了基于Stacking的三层检测模型,通过多个弱学习器构成一个强学习器提高检测模型性能.实验结果表明,提出的特征提取方法可以获得更关键的特征,设计的检测模型的准确率、精确率、召回率均优于单一模型且具有良好的泛化性,证明了检测方法的有效性.

In many IoT scenarios, the resources of terminal devices are limited, and it is difficult to provide services with low latency and low energy consumption. Mobile edge computing is an effective solution by offloading computing tasks to edge server processing. There are some problems in the existing offloading decision algorithms: the offloading decision method based on heuristic algorithms cannot dynamically adjust the policy in the changing environment; the offloading algorithm based on deep reinforcement learning will lead to slow convergence and poor exploration effect due to the problem of dimension explosion. To solve the above problems, this paper designs an offloading decision algorithm to make dynamic decisions in a mobile edge computing network with multi-device access. The algorithm comprehensively considers the energy consumption of terminal equipment, offloading overhead, average delay and success rate of task completion, aiming to achieve the highest total revenue of the whole system in a period of time. In this work, the online offloading problem is abstracted as a Markov decision process. Based on the Double Dueling Deep Q-Network (D3QN) algorithm, the offloading decision is designed to adapt to the highly dynamic environment of the edge computing network and solve the problem of high state space complexity. In addition, this paper innovatively introduces a clustering algorithm into deep reinforcement learning (DRL) to preprocess the action space and solve the explosion problem of the action space dimension caused by the increase of terminal devices. The experimental results show that the proposed algorithm is superior to the baseline strategies such as Deep Q-Network (DQN) algorithm in convergence speed and total reward.

The rapidly growing encrypted traffic hides a large number of malicious behaviours. The difficulty of collecting and labelling encrypted traffic makes the class distribution of dataset seriously imbalanced, which leads to the poor generalisation ability of the classification model. To solve this problem, a new representation learning method in encrypted traffic and its diversity enhancement model are proposed, which uses the diversity of images to represent the diversity of traffic samples. First, the encrypted traffic is transformed into Markov images. Then, a diversity maximisation Markov-GAN based on the Simpson index is designed to generate new Markov images. Finally, the balanced Markov image set is sent to the CNN for classification. Experimental results show that the proposed method can predict the whole dataset space with only a few original samples. And the classification accuracies under different imbalance degrees are significantly improved, all of which are over 90%. The enhanced Markov image set can effectively alleviate performance generalisation deviation caused by different network depths. Even an ordinary CNN has almost the same classification effect as VGG13 and VGG16. Compared with other data enhancement methods, the Markov-GAN only needs to balance the transform domain dataset, which is lightweight, easy to train and has stronger amplification ability.

With the rapid development of Android, a major mobile Internet platform, Android malware attacks have become the number one threat to mobile Internet security. Traditional malware detection methods have low precision and greater time complexity. At present, image detection methods based on deep learning are used in malware detection. However, most of these methods are based on the largescale convolutional neural network model (such as VGG16). The computation and weight files of these models are very large, so they are not suitable for mobile Internet platforms with limited computation. A novel detection method based on a lightweight convolutional neural network is presented in this study. It transforms Android malware classes.dex, Androidmanifest.xml, and resource.arsc into RGB images and uses the lightweight convolutional neural network to extract the features of RGB images automatically. The experimental results of this study indicate that the method performs well in terms of precision and speed of detection.

In the intelligent era of human-computer symbiosis, the use of machine learning method for covert communication confrontation has become a hot topic of network security. The existing covert communication technology focuses on the statistical abnormality of traffic behavior and does not consider the sensory abnormality of security censors, so it faces the core problem of lack of cognitive ability. In order to further improve the concealment of communication, a game method of “cognitive deception” is proposed, which is aimed at eliminating the anomaly of traffic in both behavioral and cognitive dimensions. Accordingly, a Wasserstein Generative Adversarial Network of Covert Channel (WCCGAN) model is established. The model uses the constraint sampling of cognitive priors to construct the constraint mechanism of “functional equivalence” and “cognitive equivalence” and is trained by a dynamic strategy updating learning algorithm. Among them, the generative module adopts joint expression learning which integrates network protocol knowledge to improve the expressiveness and discriminability of traffic cognitive features. The equivalent module guides the discriminant module to learn the pragmatic relevance features through the activity loss function of traffic and the application loss function of protocol for end-to-end training. The experimental results show that WCCGAN can directly synthesize traffic with comprehensive concealment ability, and its behavior concealment and cognitive deception are as high as 86.2% and 96.7%, respectively. Moreover, the model has good convergence and generalization ability and does not depend on specific assumptions and specific covert algorithms, which realizes a new paradigm of cognitive game in covert communication.

针对物联网设备很容易被攻击者利用来入侵网络的问题,设计实现了一种将基于网络拓扑改组的移动目标防御(network topology shuffling-based moving target defense,NTS-MTD)和网络欺骗相结合的物联网集成防御机制,并基于该防御机制设计了一种诱饵路径优化算法(decoy path-based optimization algorithm,DPOA)来进行网络拓扑改组优化.在软件定义网络(software defined network,SDN)的支持下主动改变真实节点和诱饵节点的网络拓扑,实现物联网环境中的网络拓扑改组.通过一个带有安全度量的图形安全模型(graphical security model,GSM)研究防御机制有效性,并使用三个指标对DPOA的安全性和性能进行衡量.仿真结果表明,基于DPOA的方案防御成本显著降低且安全性高,更适应于物联网.

核心提示 深圳市应用多源时空大数据和三维可视化等技术,构建居住空间布局选址智慧系统,为居住空间布局选址工作提供了全流程、全要素的决策辅助,有力支撑了居住空间相关研究和业务. 深圳市经济社会活动高度集中,空间要素类型多、流向复杂,增加居住用地供给、提升居住空间品质难度很大.传统的居住空间布局选址工作主要基于法定图则、用地调查、土地收储等数据,依靠人工进行地块筛选和空间判断,数据类型有限、自上而下特征明显,不能有效反映居住空间的现状,且重复性工作多、工作效率较低[1].当前,以大数据为代表的新一代信息技术快速发展,各类空间数据的获取、存储、分析、表现能力显著提高.因此,有必要发挥新技术优势,构建覆盖居住空间布局选址全流程、全要素的智慧系统,高效、深入地开展居住空间现状分析和需求测算.

"新工科"背景下的创新工程教育,对高校课程教学改革提出了更高的要求."电工学"课程是工科非电类专业的技术基础课程,对培养学生工程能力影响重大.本文介绍了"新工科"背景下的"电工学"课程的教学理念和目标,并对教学内容和形式进行了探讨.