Top PDF Dynamic Model Generation and Classification of Network Attacks

Dynamic Model Generation and Classification of Network Attacks

Dynamic Model Generation and Classification of Network Attacks

When attempting to read malicious network traffic, security analysts are challenged to determine what attacks are happening in the network at any given time. This need to analyze data and attempt to classify the data requires a large amount of manual time and knowledge to be successful. It can also be difficult for the analysts to determine new attacks if the data is unlike anything they have seen before. Because of the ever- changing nature of cyber-attacks, a need exists for an automated system that can read network traffic and determine the types of attacks present in a network. Many existing works for classification of network attacks exist and contain a very similar fundamental problem. This problem is the need either for labeled data, or batches of data. Real network traffic does not contain labels for attack types and is streaming packet by packet. This work proposes a system that reads in streaming malicious network data and classifies the data into attack models while dynamically generating and reevaluating attack models when needed.
Show more

64 Read more

Network Attacks Detection by Hierarchical Neural Network

Network Attacks Detection by Hierarchical Neural Network

e) Optimizing the number of the middle layer neurons: In order to optimize the number of the neurons of the middle layer, we performed the Training and Test- ing operations with neurons of different numbers. Then we selected the number of neurons of the state in which we observed the minimum testing data clas- sification error as the optimal number of neurons. So, the optimal number of neurons will be based on the Table 3. We have performed the Training and Test- ing operation with neurons of different numbers over 50 times before selecting this optimal number of neurons. Its worth mentioning that the rate of the vari- ability (uncertainty) of the neural network classification is in the interval of its two sequential performances. Therefore, we could say that its so important to know the efficient parameters in the neural network performance and to confront them expertly
Show more

14 Read more

A Threat Model Approach for Classification of Network Layer Attacks in WSN

A Threat Model Approach for Classification of Network Layer Attacks in WSN

Based on the location of attackers, the network attacks in WSN can be categorized as outsider or insider i.e. external or internal respectively. It is based on whether the attacker is a legitimate node of the network or is not a part of the network. If the intruding node is not an authorized participant of the sensor network it can be used to launch passive attacks. In such cases, the attacker has no special access to the sensor network. Whereas an inside attacker or internal threat is an authorized participant in the sensor network which has gone hostile [4]. Insider attacks may be mounted from either compromised sensor nodes running malicious code or adversaries who have stolen the key material, code, and data from legitimate nodes, and who then use one or more laptop- class devices to attack the network. Threats that are external may cause passive eavesdropping on data transmissions. They may also extend to injecting bogus data into the network so that network resources are consumed and then raise Denial of Service (DoS) attack. To prevent such attacks the best methods are the authentication and encryption techniques that shall prevent such attackers from gaining any special access to the network. Table 3 defines the various functions and effects of external based (Outside) attacks. Table 4 defines the various functions and effects of internal based (Inside) attacks [13].
Show more

6 Read more

Switching Model of a Dynamic Social Network

Switching Model of a Dynamic Social Network

[21] K. S. Park, B. H. Cho, D. H. Lee, S. H. Song, J. S. Lee, Y. J. Chee, I. Y. Kim, and S. I. Kim, “Hierarchical support vector machine based heartbeat classification using higher order statistics and hermite basis function,” 2008 Comput. Cardiol., pp. 229–232, 2008. [22] T. Lahiri, S. Sarkar, S. Sanyal, A. A. Morozov, and Y. V Obukhov, “Clustering of Signal Components within Most Likely ECG Episodes to Analyze the ECG-Waves 1,” vol. 19, no. 1, pp. 30–34, 2009.

6 Read more

Convolutional neural network on three orthogonal planes for

dynamic texture classification

Convolutional neural network on three orthogonal planes for dynamic texture classification

In this paper, we developed a new approach designed for the analysis of DT sequences based on CNNs applied on three orthogonal planes. We have shown that training independent CNNs on three orthogonal planes and combining their outputs in an ensemble model manner performs well on DT classification by learning to jointly recognize spatial and dynamic patterns. We based this work on our previously described T-CNN (specifically designed for texture images) and developed a new network for small images (Dyntex++ and UCLA) as well as deeper T-CNN networks adapted from GoogleNet to texture analysis. We experimented with our approach on the most used DT datasets in the literature, yet a major problem remains the lack of larger and more challenging DT datasets to fully exploit the power of deep learning. Our deepest approach (DT-GoogleNet) obtains slightly higher accuracy than the shallower one (DT-AlexNet) and established a new state of the art on the DynTex and Dyntex++ databases. It achieved less than 1% lower accuracy than the state of the art on the UCLA database which contains fewer training samples, making the CNN training difficult. The neural network learning process enables our approach to analyze and learn from many diverse datasets and setups with a good invariance to rotation, illumination, scale, sequence length and camera motion. Thus our method obtains high accuracy in all the tested datasets ( > 98%) whereas previous methods in the literature are more specialized in the analysis of one particular database.
Show more

19 Read more

Network Simulated Generation of Human Faces with Expressions and Orientations for Deep Learning Classification

Network Simulated Generation of Human Faces with Expressions and Orientations for Deep Learning Classification

self-report and observer measurements are extremely subjective. Nevertheless, patients may suffer from various types of pain. Asking patients to describe their pain may not be sufficient to guide medical staff to suitable interventions for specific types of pain. Currently, several techniques can be employed to assess pain, such as pathological examination, neurological examination, or imaging technologies, but they can be invasive or expensive for patients. Thus, for screening of pain [2], some researchers have proposed facial expression as a potential solution for pain detection. The most standard database used to measure pain algorithms is the UNBC-McMaster shoulder pain expression archive. Human face data have been used to classify and recognize pain in this system, which resembles an array of faces that detects moving faces and posture of humans. Patients are shown these images when they receive diagnosis. In the past [3], artificial neural networks used for image detection or security systems can achieve that goal in several years, and several enhancements in different duties. Deep learning is an algorithm that is an improvement from the artificial neural network. Nevertheless, deep learning exhibits disadvantages: It requires a considerable amount of data, and the data must be manually labeled to function correctly. The facial expression of humans is apparently a sensitive, limited resource. For example, it is difficult to predict the shape of the faces in expressions and orientations. Furthermore, the collecting method requires crystal clear pictures. Hence, a generative adversarial network (GAN) model is used to generate image data, which are used for training and testing. GANs constitute one of the most popular topics in deep learning that has been proposed by Goodfellow et al. [4]. Currently, some researchers have employed GAN for performing several things, including flower models or an MNIST database. It functions similar to neural networks, which comprise two deep neural networks: the generator and discriminator [5]. The generator can generate fake images that the discriminator attempts to distinguish with real images [6]. It continues this process until the discriminator cannot classify between the real and generated images. In this study, a method that improves the performance of deep learning for classification and recognition via the generation increased amount of data using GAN is proposed. Fig. 1 shows the structure of the proposed GAN combined with deep learning. The facial expressions of patients with shoulder pain and several postures are focused. The real images are programmed in GAN during the training process. Next, GAN generates
Show more

8 Read more

SGM: Sequence Generation Model for Multi label Classification

SGM: Sequence Generation Model for Multi label Classification

In recent years, some neural network models have also been used for the MLC task. Zhang and Zhou (2006) propose the BP-MLL that utilizes a fully-connected neural network and a pairwise ranking loss function. Nam et al. (2013) propose a neural network using cross-entropy loss instead of ranking loss. Benites and Sapozhnikova (2015) increase classification speed by adding an extra ART layer for cluster- ing. Kurata et al. (2016) utilize word embeddings based on CNN to capture label correlations. Chen et al. (2017) propose to represent semantic information of text and model high-order label correlations by combining CNN with RNN. Baker and Korhonen (2017) initialize the final hidden layer with rows that map to co-occurrence of labels based on the CNN architecture to improve the performance of the model. Ma et al. (2018) propose to use the multi-label classification algorithm for machine translation to handle the situation where a sentence can be translated into more than one correct sentences.
Show more

12 Read more

A Research Study on Accountability of Dynamic IP 
                      Generation in a Campus Network (Wi-Fi)

A Research Study on Accountability of Dynamic IP Generation in a Campus Network (Wi-Fi)

Hence, knowledge of such address block/pool can assist network operators/security analyst to suspicious activities on these blocks, detecting and preventing attacks from inside hosts. The behavior of network is also established with the knowledge of dynamic and static addresses with appropriate hosts for anomaly detection and behavior tracking. Whether IP addresses are dynamic or static is not be available on the network, even for those within one’s own network. This is particularly true for large networks with decentralized management, where large block of addresses are allocated and delegated to sub organizations which control and managed, How these address are assigned and utilized, while it is possible to defer whether an IP address is dynamic or static by its DNS name, such an approach may not always be feasible not accurate for a variety of reason:
Show more

6 Read more

An Ensemble Model for Classification of Attacks with Feature Selection based on KDD99 and NSL KDD Data Set

An Ensemble Model for Classification of Attacks with Feature Selection based on KDD99 and NSL KDD Data Set

Now a days the rapid development and popularity of Internet and Intranet, the security is very important for network. IDS is an emerging area of research in computer security and network with growing usages of Internet and Intranet in everyday life. IDS can identify the user’s activity as either normal or anomaly (Intrusion) and protect system for unauthorized users or attackers .There are various techniques applied by different authors to develop an Intrusion Detection System (IDS) in which data mining technique is one of the most widely used for classification of data. Li, Y. et al. [6] have applied various feature reduction method on KDD99 data set. In case of Gradually Feature Reduced (GFR) with 19 features, Support Vector Machine (SVM) classifier achieved high accuracy with 98.62% for intrusion detection. Koc, L. et al. [5] have introduced Hidden Naive Bayes (HNB) model with promotional k-interval discretization and INTERACT feature selection method to develop IDS. They have compared proposed model with traditional Naive Bayes methods. Our proposed model gives satisfactory result with 93.72% of accuracy in multiclass classification problem for intrusion detection in case of KDDCUP99 data set. Altwaijry, H. et al. [7] have suggested Bayesian network to improve the accuracy of R2L type of attack.
Show more

6 Read more

A dynamic power law sexual network model of gonorrhoea outbreaks

A dynamic power law sexual network model of gonorrhoea outbreaks

Human networks of sexual contacts are dynamic by nature, with partnerships forming and breaking continuously over time. Sexual behaviours are also highly heterogeneous, so that the number of partners reported by individuals over a given period of time is typically distrib- uted as a power-law. Both the dynamism and heterogeneity of sexual partnerships are likely to have an effect in the patterns of spread of sexually transmitted diseases. To represent these two fundamental properties of sexual networks, we developed a stochastic process of dynamic partnership formation and dissolution, which results in power-law numbers of part- ners over time. Model parameters can be set to produce realistic conditions in terms of the exponent of the power-law distribution, of the number of individuals without relationships and of the average duration of relationships. Using an outbreak of antibiotic resistant gonor- rhoea amongst men have sex with men as a case study, we show that our realistic dynamic network exhibits different properties compared to the frequently used static networks or homogeneous mixing models. We also consider an approximation to our dynamic network model in terms of a much simpler branching process. We estimate the parameters of the generation time distribution and offspring distribution which can be used for example in the context of outbreak reconstruction based on genomic data. Finally, we investigate the impact of a range of interventions against gonorrhoea, including increased condom use, more frequent screening and immunisation, concluding that the latter shows great promise to reduce the burden of gonorrhoea, even if the vaccine was only partially effective or applied to only a random subset of the population.
Show more

20 Read more

A dynamic model of mobile telephony subscription incorporating a network effect

A dynamic model of mobile telephony subscription incorporating a network effect

Several institutional developments have had an impact on mobile telephony growth. 1 Introduction of digital technology substantially relaxed the radio spectrum constraint. Transmission rates increased from 0.33 bits/s/Hz to approximately 1.40 bits/s/Hz as TDMA allowed more efficient use of the radio spectrum (Gruber & Valletti, 2003). Digital technology brought features for commercial mobile telecommunications not available with analogue technology. In particular, digital technology permits data transmission, e.g., short message service, e-mail and increased sound quality. Digital networks require lower power levels for operation that results in lighter and smaller handsets (ITU, 1999). Second-generation (2G) mobile systems operation have drawn on the experience of first-generation (FG) mobile systems in the realization of network effects and economies of scale. This has resulted in the creation of fewer systems than for FG mobile.
Show more

13 Read more

Packet Classification for Preventing Selective Jamming Attacks

Packet Classification for Preventing Selective Jamming Attacks

This paper considers the problem of an attacker disrupting an encrypted victim wireless ad hoc network through jamming. Jamming is broken down into layers and this paper focuses on jamming at the Transport/Network layer. Jamming at this layer exploits AODV and TCP protocols and is shown to be very effective in simulated and real networks when it can sense victim packet types, but the encryption is assumed to mask the entire header and contents of the packet so that only packet size, timing, and sequence is available to the attacker for sensing. A sensor is developed that consists of four components. The first is a probabilistic model of the sizes and inter-packet timing of different packet types. The second is a historical method for detecting known protocol sequences that is used to develop the probabilistic models, the third is an active jamming mechanism to force the victim network to produce known sequences for the historical analyzer, and the fourth is the online classifier that makes packet type classification decisions. The method is tested on live data and found that for many packet types the classification is highly reliable. The relative roles of size, timing, and sequence are discussed along with the implications for making networks more secure.
Show more

12 Read more

An Efficient Decision Tree Model for Classification of Attacks with Feature Selection

An Efficient Decision Tree Model for Classification of Attacks with Feature Selection

feature validity based reduction method (FVBRM) applied on one of the efficient classifier Naive Bayes on reduced data set with 24 features for intrusion detection. Result obtained in this case is better as compare to case based feature selection (CFS), gain ratio (GR), info gain ratio (IGR) to design efficient and effective network intrusion detection system. Y., Li et al.[5] have applied various feature reduction methods on KDD99 data set. They have obtained 98.62% accuracy using gradually feature reduction technique with 19 features through support vector machine and 10-fold cross validation. Koc, L. et al. [4] have introduced Hidden Naive Bayes (HNB) model with promotional k-interval discretization and INTERACT feature selection method. They have compared their proposed model with traditional Naive Bayes method. A recent literature by Ibrahim, Laheeb M. et al [3] focuses on self organization map (SOM) model which compare detection rate in between two data sets: KDD99 and NSL-KDD. Detection rate of SOM with KDD99 is 92.37% while it is 75.49% for NSL-KDD data. Some other authors have worked on binary classification problem which can classify data into two class like normal and attack. Mrutyunjaya Panda et al. [8] have suggested hybrid technique with combination of random forest, dichotomies, and ensembles of balanced nested dichotomies (END) for binary class problem, which gives detection rate 99.50% and low false alarm rate of 0.1%. They have evaluated the performance of model with other measures like F-value, precision and recall. There are various authors who have worked on various techniques and applied feature selection techniques as one of the important component. Literature review revealed that feature selection is one of the most essential parts of development of IDS.
Show more

7 Read more

Web Text Classification Using Genetic Algorithm and a Dynamic Neural Network Model

Web Text Classification Using Genetic Algorithm and a Dynamic Neural Network Model

The general philosophy of the dynamic neural network model is based upon the principle of learning and accumulating knowledge at each layer, propagating and adjusting this knowledge forward to the next layer, and repeating these steps until the desired network performance criteria are reached. As in classical neural networks, the dynamic neural network architecture is composed of an input layer, hidden layers and an output layer. The input layer accepts external data to the model. In dynamic neural network, unlike classical neural networks, the number of hidden layers is not fixed a priori. They are sequentially and dynamically generated until a level of performance accuracy is reached. Additionally, the proposed approach uses a fixed number of hidden nodes (four) in each hidden layer. This structure is not arbitrary, but justified by the estimation approach. At each hidden layer, the network is trained using all observations in the training set simultaneously, so as to minimize a stated training accuracy measure such as mean
Show more

7 Read more

Attentive Temporal Pyramid Network for Dynamic Scene Classification

Attentive Temporal Pyramid Network for Dynamic Scene Classification

To achieve more accurate recognition of dynamic scene, it would be beneficial to fully explore the temporal dynamic information. There are basically two pathways to model temporal clues within CNNs. One way is to explicitly model the video as an ordered sequence of frames based on long short-term memory (LSTM) (Donahue et al. 2015) or gated recurrent unit (GRU) (Chung et al. 2014). These models usu- ally adopt memory cells to store, modify and access internal state so as to discover the long-range sequential informa- tion. Alternatively, another way of capturing the temporal information in CNNs resorts to the two-stream architecture (Simonyan and Zisserman 2014a) which uses both RGB and dense optical flows as the inputs for CNNs. By incorporating these two sources of information, the model encodes both spatial and temporal clues in the two-stream network. De- spite the success of these methods, the computational cost tends to be high, and in addition, indiscriminately using en- tire video frames for modeling will introduce negative ef- fects of irrelevant and noisy frames, thereby compromising the classification performance.
Show more

8 Read more

A Generative Attentional Neural Network Model for Dialogue Act Classification

A Generative Attentional Neural Network Model for Dialogue Act Classification

In this work, we have proposed a new gated at- tention mechanism and a novel HMM-like con- nection in a generative model of utterances and dialogue acts. Our experiments show that these two innovations significantly improve the accu- racy of DA classification on the MapTask and Switchboard corpora. In the future, we plan to apply these two innovations to other sequence-to- sequence learning tasks. Furthermore, DA classi- fication itself can be seen as a preprocessing step in a dialogue system’s pipeline. Thus, we also plan to investigate the effect of improvements in DA classification on the downstream components of a dialogue system.
Show more

6 Read more

A Semi Supervised Bayesian Network Model for Microblog Topic Classification

A Semi Supervised Bayesian Network Model for Microblog Topic Classification

In this paper, we proposed a novel scheme to classify microblogging messages, which addresses three concerns in microblog classifications. First, the incorporation of external resources to supple- ment the short microblogs well compensates the data sparseness issue. Second, the semi-supervised classifier seamlessly fuse labeled data structure and external resources into the training process, which reduced the requirement for manually labeling to a certain degree. Third, we model the cate- gory probability of a given message based on the category-word distribution, and this successfully avoided the difficulty brought about by the spelling errors that are common in microblogging mes- sages. We proposed a semi-supervised learning approach to classify microblogging messages, and the experimental results demonstrated its effectiveness as compared to existing the state-of-the-art methods, as well as practically extension to large-scale dataset.
Show more

16 Read more

Mobile Robot  Dynamic Model Controlling using Wavelet Network

Mobile Robot Dynamic Model Controlling using Wavelet Network

In this paper, mobile robot control system is implemented and tested by using matlab/simulink software package version 7.12.0 (R2011a), and the simulation results reveal that the method of using wavelet neural network as a motion controller is feasible because the wavelet neural network is succeeded to make the actual trajectory quickly approach to the desire trajectory in a short time. And the deviation between the actual and desire trajectories is very small. Also, the PSO algorithm is succeeded to optimize the wavelet neural network parameters by depending on the mean square error value.
Show more

5 Read more

Dynamic Simulation Model for Gas Transmission Pipeline Network System

Dynamic Simulation Model for Gas Transmission Pipeline Network System

The traditional hydraulic simulation model is very complex. A set of equations is used to describe the hydraulic model of a whole station. The equations are usually non-linear. The calculation amount of solving the equations increases exponentially with the complexity of the pipeline network [3]. When the pipeline network is complex to a certain extent, it is impossible to realize real-time pressure calculation [4-5].

8 Read more

Survey on Classification of Attacks and Security Mechanism in Wireless Network

Survey on Classification of Attacks and Security Mechanism in Wireless Network

Security is the process of preventing and detecting unauthorized use of wireless computer. Prevention measures help you to stop unauthorized users from accessing any part of your computer system. Detection helps you to determine whether or not someone attempted to break into wireless network system, if they were successful, and what they may have done. X. Liang [10], [14] propose some mutual authentication and key exchange methods in wireless network for secure communication. In [13] and [14], public key cryptography such as digital signature and Diffie – Hellman key exchange, is accepted on the basis of SC- based Schemes, which can further improve the security of Wireless Service. Mainly existing wireless schemes for secure communication in network can mainly be classified into three categories: symmetric- cryptosystem-based (SC based), asymmetric- cryptosystem-based (AC-based), and hybrid schemes. The EAP-based authentication and key agreement protocols [8],[16],can also be called as SC-based secure wireless method are designed based on standard protocols. SC-based methods are widely used because they are well match with accepted protocols.
Show more

6 Read more

Show all 10000 documents...