检索结果-内蒙古大学图书馆

Network Protocol Recognition Based on Convolutional Neural Network

学校读者我要写书评

暂无评论

China Communications 2020年第4期17卷 125-139页

作者： Wenbo Feng Zheng Hong Lifa Wu Menglin Fu Yihao Li Peihong Lin Institute of Command and Control Engineering Army Engineering University of PLANanjing210007China School of Computer Science and Technology Nanjing University of Posts and TelecommunicationsNanjing210023China

How to correctly acquire the appropriate features is a primary problem in network protocol recognition *** to avoid the trouble of artificially extracting features in traditional methods and improve recognition accuracy,a network protocol recognition method based on Convolutional Neural Network(CNN)is *** method utilizes deep learning technique,and it processes network flows ***,normalization is performed on the intercepted network flows and they are mapped into two-dimensional matrix which will be used as the input of ***,an improved classification model named Ptr CNN is built,which can automatically extract the appropriate features of network ***,the classification model is trained to recognize the network *** proposed approach is compared with several machine learning *** results show that the tailored CNN can not only improve protocol recognition accuracy but also ensure the fast convergence of classification model and reduce the classification time.

关键词： convolutional neural network protocol recognition network flow classification model

AndrOpGAN: An Opcode GAN for Android Malware Obfuscations 1

学校读者我要写书评

暂无评论

3rd International Conference on Machine Learning for Cyber Security, ML4CS 2020

作者： Zhang, Xuetao Wang, Jinshuang Sun, Meng Feng, Yao Institute of Command Control Engineering Army Engineering University of PLA NanjingJiangsu210007 China

ISBN: (数字)9783030622237

ISBN: (纸本)9783030622220

With the rapid development of Android platform, the number of Android malwares is growing rapidly. Due to the limitations of traditional static and runtime Android malware analysis methods, machine learning based approaches are widely adopted recently. Whereas, evading methods are also emerging, e.g. data set pollution, feature modification. Current feature modifications are mainly based on high-level features such as API calls or sensitive permissions. Our contribution is to show it is also feasible to deceive the detectors by modifying underlying features. Through this confusion, detector deceiving can be achieved. An Android malware opcode distribution feature modification system AndrOpGAN was proposed. To adjust the opcode distribution of malware, Deep Convolution Generative Adversarial Networks (DCGAN) was proposed to generate opcodes distribution features, and opcodes would be inserted through an Opcode Frequency Optimal Adjustment algorithm (OFOA). OFOA module can keep the APK running normally after insertion with a low modification cost. Test results against four detectors show that more than 99% APKs processed by AndrOpGAN could bypass detections successfully. Test results against VirusTotal shows that, the number of successful detection engines decreased 20%–44%. AndrOpGAN validates the feasibility of such attacks based on underlying feature modifications and provides a prototype system for researchers to improve detector’s performance. © 2020, Springer Nature Switzerland AG.

关键词： Android malware

A New Adjacency Matrix Configuration in GCN-based Models for Skeleton-based Action Recognition

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Fang, Zheng Zhang, Xiongwei Cao, Tieyong Zheng, Yunfei Sun, Meng Institute of Command and Control Engineering Army Engineering University Jiangsu Nanjing210001 China The Army Artillery and Defense Academy of PLA Jiangsu Nanjing211131 China

Human skeleton data has received increasing attention in action recognition due to its background robustness and high efficiency. In skeleton-based action recognition, graph convolutional network (GCN) has become the mainstream method. This paper analyzes the fundamental factor for GCN-based models - the adjacency matrix. We notice that most GCN-based methods conduct their adjacency matrix based on the human natural skeleton structure. Based on our former work and analysis, we propose that the human natural skeleton structure adjacency matrix is not proper for skeleton-based action recognition. We propose a new adjacency matrix that abandons all rigid neighbor connections but lets the model adaptively learn the relationships of joints. We conduct extensive experiments and analysis with a validation model on two skeleton-based action recognition datasets (NTURGBD60 and FineGYM). Comprehensive experimental results and analysis reveals that 1) the most widely used human natural skeleton structure adjacency matrix is unsuitable in skeleton-based action recognition;2) The proposed adjacency matrix is superior in model performance, noise robustness and transferability. Copyright © 2022, The Authors. All rights reserved.

关键词： Musculoskeletal system

Research on Autonomous Maneuvering Strategies of UAV Swarm for Saturation Attack

学校读者我要写书评

暂无评论

Research on Autonomous Maneuvering Strategies of UAV Swarm f...

2022 IEEE International Conference on Unmanned Systems, ICUS 2022

作者： Zhang, Tingting Zuo, Yi Zhang, Dongning Ding, Fei Huang, Jun Lin, Yun College of Command and Control Engineering Army Engineering University of the Pla Nanjing China Key Laboratory of Information System Requirements Nanjing China Fifty-fourth Research Institute of China Electronics Technology Group Corporation Shijiazhuang China Internet of Things Academy Nanjing University of Posts and Telecommunications Nanjing China Jsti Group Nanjing China Harbin Engineering University Department of Information and Communication Engineering Harbin China

ISBN: (数字)9781665484565

ISBN: (纸本)9781665484565

In order to solve the problem of unmanned aerial Vehicle (UAV) cluster saturation attack, the autonomous maneuver strategy of UAV is studied. Based on the distributed partial observable Markov decision process, the autonomous maneuver decision model of UAV is constructed, and the Recurrent Multi-Agent Deep Deterministic Policy Gradient (RMADDPG) is used to learn the maneuver strategy of UAV cluster saturation attack. Combined with the typical characteristics of saturation attack scenario, two reward functions of global reward and local reward for UAV maneuver are designed. The trained UAVs can effectively perform suicide saturation attack task. According to the mission target design evaluation index of saturation attack, the Monte Carlo analysis method is used to compare the proposed method with particle swarm optimization (PSO) algorithm in three aspects: inter aircraft collision avoidance, NFZ avoidance and simultaneous attack. The experimental results show that the trained UAVs can effectively perform the suicide saturation attack task under the condition of a small amount of communication, which provides a method for the application of multi-agent deep reinforcement learning to UAV cluster operations. © 2022 IEEE.

关键词： Unmanned aerial vehicles (UAV)

GAIL-PT: A Generic Intelligent Penetration Testing Framework with Generative Adversarial Imitation Learning

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Chen, Jinyin Hu, Shulong Zheng, Haibin Xing, Changyou Zhang, Guomin Institute of Cyberspace Security Zhejiang University of Technology Hangzhou China College of Information Engineering Zhejiang University of Technology Hangzhou China College of Command and Control Engineering Army Engineering University of PLA Nanjing China

Penetration testing (PT) is an efficient network testing and vulnerability mining tool by simulating a hacker’s attack for valuable information applied in some areas, e.g., the host’s operating and database systems. Compared with manual PT, intelligent PT has become a dominating mainstream due to less time-consuming and lower labor costs. Among the already proposed intelligent PT methods, reinforcement learning (RL) based PT has achieved state-of-the-art (SOTA) performance. Unfortunately, RL-based PT is still challenged in real exploitation scenarios because the agent’s action space is usually high-dimensional discrete, thus leading to algorithm convergence difficulty. Besides, most PT methods still rely on the decisions of security experts. Addressing the challenges, for the first time, we introduce expert knowledge to guide the agent to make better decisions in RL-based PT, and propose a Generative Adversarial Imitation Learning based generic intelligent Penetration Testing framework, denoted as GAIL-PT, to solve the problems of higher labor costs due to the involvement of security experts and high-dimensional discrete action space. Specifically, first, we manually collect the state-action pairs to construct an expert knowledge base when the pre-trained RL / DRL model executes successful penetration testings. Second, we input the expert knowledge and the state-action pairs generated online by the different RL / DRL models into the discriminator of GAIL for training. At last, we apply the output reward of the discriminator to guide the agent to perform the action with a higher penetration success rate to improve PT’s performance. Extensive experiments conducted on the real target host and simulated network scenarios show that GAIL-PT achieves SOTA penetration performance against DeepExploit in exploiting actual target Metasploitable2 and Q-learning in optimizing penetration path, not only in small-scale with or without honey-pot network environments, but also in th

关键词： Reinforcement learning

Research on Behavioral Decision Reward Mechanism of Unmanned Swarm System in Confrontational Environment 7th

学校读者我要写书评

暂无评论

Research on Behavioral Decision Reward Mechanism of Unmanned...

7th International Conference on Artificial Intelligence and Security, ICAIS 2021

作者： Zhang, Tingting Lan, Yushi Song, Aiguo School of Command and Control Engineering Army Engineering University Nanjing210017 China The 28th Research Institute of China Electronics Technology Group Corporation National Defence Science and Technology Key Laboratory Nanjing210017 China School of Instrumental Science and Engineering Southeast University Nanjing210096 China

ISBN: (纸本)9783030786144

Unmanned swarm system is regarded as a multi-agent system. At present, many kinds of research use deep reinforcement learning algorithms to solve the problem of multi-agent autonomous behavioral decision. The multi-agent deep deterministic policy gradient (MADDPG) algorithm proposed by Google’s DeepMind team pioneered the application of multiple agents in collaborating and confronting complex scenarios, and can be used to solve the problem of generating an autonomous strategy for unmanned swarm in continuous action space. Experiments show that this algorithm has an unsatisfactory convergence and stability in a competitive environment. The key to multi-agent strategy optimization is the design of the reward function. Thus, the paper improves the algorithm of the reward model for multi-agent behaviors in the confrontational environment. First, the advantages of the MADDPG algorithm for implementing multi-agent behavioral decision in a confrontational environment and the training and execution framework of the reward function are introduced. Second, a per-distance reward mechanism and the addition of distance parameters are proposed to improve the efficiency of the agents’ behavioral decision and solve the delayed reward problem. The global sharing value reward strategy is changed to the experience pool sharing value reward strategy to avoid the phenomenon of lazy agents and improve the efficiency of the agents’ cooperation. Finally, after placing the improved reward mechanism into the MADDPG algorithm, experimental results show that the improved reward function has improved convergence, the algorithm training cycle is shortened, the function curve stabilizes faster, and multiple agents have better performance in the confrontational environment. This paper provides reference methods for intelligent behavioral decisions of the unmanned swarm system. © 2021, Springer Nature Switzerland AG.

关键词： Efficiency

Estimating Cosmological Parameters and Reconstructing Hubble Constant with Artificial Neural Networks: A Test with covariance matrix and mock H(z)

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Chen, Jie-Feng Zhang, Tong-Jie He, Peng Zhang, Tingting Zhang, Jie College of Physics and Electronic Engineering Qilu Normal University Jinan250200 China Institute for Frontiers in Astronomy and Astrophysics Beijing Normal University Beijing102206 China School of Physics and Astronomy Beijing Normal University Beijing100875 China Bureau of Frontier Sciences and Basic Research Chinese Academy of Sciences Beijing100190 China College of Command and Control Engineering PLA Army Engineering University Nanjing210000 China

In this work, we present a new approach to estimate the cosmological parameters and reconstruct the Hubble constant. We reconstructed the function from observational Hubble data using an Artificial Neural Network. The training data we used are covariance matrix and mock H(z). With the reconstructed H(z), we can get the Hubble constant, and thus do the comparison with the CMB-based measurements. In order to constrain the cosmological parameters, we sampled data points from the reconstructed data and estimated the posterior distribution. Furthermore, we did many comparisons to test the quality of the reconstructed data. Finally, with the result of the test, we propose that the H(z) reconstructed by our artificial neural network can represent the actual distribution of the real observational data, and can be used in further cosmological research. © 2024, CC BY-NC-SA.

关键词： Cosmology

Effect of Neutrinos on Angular Momentum of Dark Matter Halo

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Chen, Yu Lu, Chang-Zhi Lu, Yu Zhang, Tingting Zhang, Tong-Jie Institute for Frontiers in Astronomy and Astrophysics Beijing Normal University Beijing102206 China Department of Astronomy Beijing Normal University Beijing100875 China Beijing Planetarium Beijing Academy of Science and Technology Beijing100044 China College of Command and Control Engineering Pla Army Engineering University Nanjing210017 China Institute for Astronomical Science Dezhou University Dezhou253023 China

Massive neutrinos are expected to affect the large-scale structure formation, including the major component of solid substances, dark matter halos. How halos are influenced by neutrinos is vital and interesting, and angular momentum (AM) as a significant feature provides a statistical perspective for this issue. Exploring halos from TianNu N-body cosmological simulation with the co-evolving neutrino particles, we obtain some concrete conclusions. First, by comparing the same halos with and without neutrinos, in contrast to the neutrino-free case, over 89.71% of halos have smaller halo moduli, over 71.06% have smaller particle-mass-reduced (PMR) AM moduli, and over 95.44% change their orientations of less than 0.65°. Moreover, the relative variation of PMR modulus is more visible for low-mass halos. Second, to explore the PMR moduli of halos in dense or sparse areas, we divide the whole box into big cubes, and search for halos within a small spherical cell in a single cube. From the two-level divisions, we discover that in denser cubes, the variation of PMR moduli with massive neutrinos decreases more significantly. This distinction suggests that neutrinos exert heavier influence on halos' moduli in compact regions. With massive neutrinos, most halos (86.60%) have lower masses than without neutrinos. Copyright © 2023, The Authors. All rights reserved.

关键词： Angular momentum

Novel Cosmological Joint Constraints in Multidimensional Observables Space with Redshift-free Inferences

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Hong, Wei Jiao, Kang Wang, Yu-Chen Zhang, Tingting Zhang, Tong-Jie Institute for Frontiers in Astronomy and Astrophysics Beijing Normal University Beijing102206 China Department of Astronomy Beijing Normal University Beijing100875 China Kavli Institute for Astronomy and Astrophysics Peking University Beijing100871 China Department of Astronomy School of Physics Peking University Beijing100871 China College of Command and Control Engineering Army Engineering University Nanjing210017 China

Cosmology constraints serve as a crucial criterion in discriminating cosmological models. The traditional combined method to constrain the cosmological parameters designates the corresponding theoretical value and observational data as functions of redshift, however, sometimes the redshift cannot be measured directly, or the measurement error is large, or the definition of redshift is controversial. In this paper, we propose a novel joint method to constrain parameters that eliminates the redshift z and makes full use of the multiple observables {F1,obs, F2,obs, · · ·, FM,obs} spanning in M-dimensional joint observables space. Considering the generality of the mathematical form of the cosmological models and the guidance from low to high dimensions, we firstly validate our method in a three-dimensional joint observables space spanned by H(z), fσ8(z) and DA(z), where the three coordinates can be considered redshift-free measurements of the same celestial body (or shared-redshift data reconstructed model independently). Our results are consistent with the traditional combined method but with lower errors, yielding H0 = 68.7 ±0.1 km s−1 Mpc−1, Ωm0 = 0.289 ±0.003, σ8 = 0.82 ±0.01 and showing alleviated parametric degeneracies to some extent. In principle, our joint constraint method allows an extended form keeping the redshift information as an independent coordinate and can also be readilydegraded to the form of a traditional combined method to constrain parameters. Copyright © 2023, The Authors. All rights reserved.

关键词： Cosmology