Review of Neural Networks Contribution in Network Security

Review of Neural Networks Contribution in Network Security

Maha Mahmood ${ }^{1}$, Belal Al-Khateeb ${ }^{2}$ and WisamMakki Alwash ${ }^{3}$
${ }^{1}$ College of Computer Science and Information Technology, University of Anbar, Ramadi, Iraq.
${ }^{2}$ College of Computer Science and Information Technology, University of Anbar, Ramadi, Iraq.
${ }^{3}$ College of Law, University of Babylon, Babel, Iraq.
maha_mahmood@computer-college.org, belal@computer-college.org, wissam_alwash@yahoo.com

Abstract

Nowadays, securing the transmitted data is the most important challenging areas of development and research in modern communication. Users are able to communicate over an insecure channel using cryptography,so an attacker cannot decrypt and understand the original message. Public key cryptography requires large computational power, huge time consumption and complexity. An Artificial Neural Network (ANN) is used in order to overcome these problems. The connection between cryptography and ANN provides a great help for the security concerns. This paper presents a review for the contribution of ANN in the field of network security

Keywords: Neural Network, Network Security, Cryptography, One Time Pad.

1 INTRODUCTION

A Neural Network (NN) is an interconnection of units or nodes,the basic element of NN is the neuron. The network processing ability is obtained by an adaptation to,or learning from, a set of training patterns.Computer security is the generic name for the automated tools designed to protect data [1]. Encryption is the most important automated tool for communications and network security. The encryption technique essentially map data to a domain in a manner that is sniffing-proof. Symmetric and asymmetric encryption are the two major techniques used in encryption. The structureof neural networks is shown in figure 1.

Figure 1: The Structure of NN.
The discrete automated control systems replaces the production process controls, which rapidly increasethe automation control systems and processes importance. This dynamic process requires an immediate information transfer to the control system, so a direct online connection is preferred. In order to have a proper working control system, there is a need for the results and feedback via communication channels that are properly working, which have the ability to transferring the essential information quickly and reliably in a safe manner.

2 COMPUTER NETWORK SECURITY AND NEURAL NETWORKS

US Department of Defines specified the trusted computer criterion and evaluation rule as the two popular network security evaluation criterion that represent the principle of the computer network security evaluation. The evaluation criterions of different countries are developed based on their actual conditions [1,5]; this

^[1]

depends on the differences in development progress and degree of the country. The computer network is considered as a very complicated system, as the network security can be affected by many factors. The evaluation of the network security is as follows [1,5]: at first the contents and scope of the network evaluation are identified and then initial analysis on the basic network conditions, security conditions and network vulnerabilities are performed. While the second step is to establish the related mathematical evaluation model then finallythe network security level is computed by using the mathematical model. The use of vector multiplication, the sigmoid function and other different approximation in order to realize self-organization, selfadaptation and self-learning. Those factors represent the major differences between NN and traditional method. The organization structureof NN is shown in Figure2.

Figure 2: The Organization Structure of NN.

3 LITERATURE REVIEW

In 2005, Khaled M. G. Noman and Hamid Abdullah Jalab [2] presented aGRNN artificial neural networks encryption system that is fixed to the secret keys. Many training iterations are tested in the proposedsystem, also the researchers tested the system with different numbers of input data and hidden neurons. The obtained results was relatively better performance than the traditional encryption methods.

In $[3,25]$ used an efficient technique that is scalable for addressing computer network security. A new asymmetric encryption mechanism based on Multi layer neural network that is trained by backpropagation learning algorithm is provided for the creation of public key and the decryption scheme. In addition, Boolean algebra is used for the creation of private key and the encryption scheme. The designed system is new and required lessmemory complexity and time. The obtained results show that the possibility of guessing keys in Data Encryption Standard method (DES) is extremely stronger than the designed system.

In 2012, InadyutiDutt, Soumya Paul and DipayanBandyopadyay [4] proposed and implemented an ANN algorithm for All-Optical-Networks (AON) security attacks. The algorithm coupling all the inputs likesource port number, destination port number, source ip address, destination ipaddress and response time into an internal node and checks the validity. If the inputs produce the desired output, then the external node is allowed to access the internal node of the network otherwise a security attack is said to have occurred on the specific node.

In 2013, Halenar Igor, JuhasovaBohuslava, Juhas Martin and Nesticky Martin [5] used the expert systems classical methods together with the neural network technologies for the insurance of data communication control system. The proposed solution defines the architecture of the neural network and its type, defines the identification of data element, the neural network input transformation solvingand examining various neural networks architecture types and activation functions together with real environment tests. The obtained results showed that the individual data packet is sufficiently classified using the designed neural network. The number of neurons in hidden layer was 20 and 40 (gives the best classification ability). In addition, other experiments with the used neural networks have shown increasing the neurons in hidden layer have no significant effect. There is a lot of work to do in the very complex area. For example, propose a complex neural networkthat can

give an ability to separate incoming packets into more classes and have the ability to search in the data stream for other anomalies.

In 2013, Navita Agarwal and Prachi Agarwal [6] used a back propagation recurrent network for implementing a finite state sequential machine. Any type of sequential machine can be used in order to train the neural network; the key representing the complexity and the level of security can do this. The obtained results are promising and opening a new researchdierctions.

In 2014 Shujan Jin [7] proposed three level and four class indicators system that are used for collecting 100 group data on the evaluation of computer network security for different scales via the scoring of an expert, optimizing NN model by using the particle swarm optimization, establishing the evaluation system model of the network security based on NN, evaluating network security and finally normalize them. The NN evaluation model is simple and practicable that eliminatesthe human being disturbance of the subjective factors. The obtained results showed that the proposed system reducedthe relative output error and improvedthe evaluation correctness rate. NN overcomes the previous evaluation methods weaknesses, improve the evaluation results precision and provide reference to prediction and control of the network security problems in future. The proposed system has some disadvantages that can be solved by using some enhancements, among those weaknesses are the computation time is too long and the converged results cannot be obtained.

In $[8,26]$ used a Deep Neural Network (DNN) for establishing an efficient Intrusion Detection System (IDS) to be used in vehicular network. The researchers used probability based feature vectors for training the DNN, those features are extracted from the in vehicular network packets by using unsupervised pre training method of deep belief networks, followed by the conventional stochastic gradient descent method. The DNN discriminates normal and hacking packets by providing the probability of each class making the system able to identify any vehicle malicious attack. The researchers comprising the mode information and the value information extracted from the network packet by a proposed novel feature vector that are efficiently used in the training and the testing. The obtained results showed that the proposed technique gives an average accurate detection ratio of $98 \%$ with a real time response to the attack and small number of layers with low computational complexity.

In 2016, S. Jegadeeswari, P. Dinadayalan and N. Gnanambigai [9] proposed a new cloud data security model that ensures high security and confidentiality in cloud data storage environment for achieving data confidentiality in the cloud database. The Dynamic Hashing Fragmented Component is used to store the fragmented sensitive data and implements cryptography on it for its security. This work is efficient for all kinds of queries when compared to conventional cloud data security models by achieving high level of data confidentiality. This work includes a cloud database that achieves high confidentiality on data security together with a Neural Data Security Model (NDSM) to ensure the high security and confidentiality in cloud data storage environment. The neural cryptographic algorithm is used in the data security model to deal with data encryption for sensitive data to enhance the confidentiality level.

In 2017, J. RubinaParveen [10] describes hacking as the mainreason of data losing in the corporate industry. Organizations oftenprotecting their data from loss or leakage by taking adequate measures. Quite often, theunexpected cyber attacks are because of the risk factors of the unchecked IT cyber security, so IDS are mainly used for securing company networks. The security solutions for business including intelligent behavioural analysis and machine learning algorithms represent a new generation technologies. In the industry, achieving one of the highest detection rates is done by using the combination of artificial intelligence and NN this is continuously demonstrated through independent tests.

In 2017, Sufyan T. Al-Janabi, Belal Al-Khateeb and Ahmed J. Abd. [11] mainly concerned in building automatic tools for various cryptanalysis tasks. This definitely requires the use of suitable intelligent techniques such as GAs and ANNs. The focus here has been on using GAs for cryptanalysis of classical ciphers and adoption of ANNs for cipher type identification. Emphasis is given to the use of GAs in cryptanalysis of classical ciphers. Another important cryptanalysis issues to be considered is cipher type detection or identification. This can be a real obstacle to cryptanalysts and it is a basic step for any automated cryptanalysis system and some other cryptanalysis tasks.

In 2018, MuriloCoutinho, Robson de Oliveira Albuquerque, Fábio Borges, Luis Javier GarcíaVillalba and TaiHoon Kim [12] protected communications with Adversarial Neural Cryptography by using secure cryptography. The researchers shown that in the right circumstances a perfectly secure cryptosystem can be learned by neural network. In addition, the obtained results demonstrated that the original adversarial neural cryptographymethodology is not good enough to achieve this goal. Moreover, the researchers presented a new CPA-ANC methodology for the purpose of improving the objective function and the learned model. In their experiments, they used simple neural networks to better understand the learned model. The obtained results showed the superiority of the proposed CPA-ANC over the original ANC methodology, as in CPA-ANC almost all the learned models were secure OTP. The researchers concluded that in order to force the solution into a strong cryptosystem, the adversary must be very powerful. In this case the standalone CPA-ANC methodology is not enough to guarantee security, the key is to design a very strong adversarial capable of breaking cryptosystems. In their model, the researchers showed that achieving this was possible; however, in general, this is a hard open problem.

In [13,27] formally checking DNNs security properties without using Satisfiability Modulo Theory SMT solvers. Instead, arithmetic interval to compute rigorous bounds on the DNN outputsis leveraged.AReluVal, a formal security analysis system for neural networks is designed, developed, and evaluated. Several novel techniques including symbolic interval arithmetic to perform formal analysis without resorting to SMT solvers are introduced. ReluVal performed 200 times faster on average than the current state-of-art solver-based approaches.The presented approach is easily parallelizable, unlike existing solver-based approaches.To minimize overestimations of output bounds, symbolic interval analysis along with several other optimizations is further presented. There are three main contributions.

• To the best of knowledge, ReluVal is the first system that leverages interval arithmetic to provide formal guarantees of DNN security.
• Naive application of interval arithmetic to DNNs is ineffective. Two optimizations - symbolic intervals and iterative refinement - that significantly improve the accuracy of interval arithmetic on DNNs are presented.
• The techniques as part of ReluVal are designed, implemented, and evaluated, the results demonstrated that it is on average 200xfasterthanReluplex, a state-of-the-art DNN verifier using a specialized solver $[15,16,17,18,19]$.

4 ADVANTAGES OF NEURAL NETWORKS IN SECURITY

The use of ANN can identify attackswhen rules are not known [9]. Patterns are recognized and recent actions happened with the usual behaviour are compared by a neural network approach,also, NN is adapted to certain constraints, in order to resolve many issues even without human intervention. Misuse are consistently detected by neural networks,also, the recognition of malicious events are improved. This makes the system enhancing flexibility against intrusions in order to be able to protect their entire organization.As a conclusion, NN identifying intrusions of in secure networks in more reliable and accurateway.

5 DISADVANTAGES OF NEURAL NETWORKS IN SECURITY

The artificial neural network depends on the data training requirement and methods in order to be able to identify intrusion. Those requirements and methods arevery critical to use and time consuming. The sequence of thousands of individual intrusion is required for training. Those individual intrusions are very sensitive to obtain and it will take time to achieve. The ‘Black box’ nature of neural network is considered as the most significant disadvantage in applying neural network to IDS. The “Black Box Problem” has affected neural networks in a number of applications [10]. This area of NN research is still ongoing.

6 CONCLUSIONS

Evolutionary computation has proven to be very useful optimization tool in many applications. Determining its efficiency as an optimization algorithm for feedforward neural network architectures. Using an automatic generation of neural network’s architecture allow it to adapt their architecture according to the selected problem that deal with it [1,14]. Today, data security is considered as a prime concern in data communication systems.ANNbased n-state sequential machine in the field of cryptography is new efficient method that can be used for encrypting and decryptingthe data.

7 FUTURE WORKS

In future, the concept of big data and deep learning is used to obtain higher cryptographic security. Also, an investigationof other evolutionary methods, like genetic programming can be applied in order to automatically generate the neural network architecture for many applications.

REFERENCES

[1] MahaMahmood andBelal Al-Khateeb, “Towards an Automatic Generation of Neural Networks”, Journal of Theoretical and Applied Information Technology, Vol.95. Issue 23, 2017.
[2] Khaled M. G. Noaman and Hamid Abdullah Jalab, “Data Security Based On Neural Network”, Task Quarterly 9 No 4, 409-414, 2005.
[3]P. Mohamed Shakeel; Tarek E. El. Tobely; Haytham Al-Feel; Gunasekaran Manogaran; S. Baskar., “Neural Network Based Brain Tumor Detection Using Wireless Infrared Imaging Sensor”, IEEE Access, 2019, vol 7, issue 1, Page(s): 1
[4] InadyutiDutt, Soumya Paul and DipayanBandyopadyay ," Security in All-Optical Network using Artificial Neural Network", International Journal of Advanced Research in Computer Science, Vol. 3, No. 2, 2012.
[5] Halenar Igor, JuhasovaBohuslava, JuhasMartin andNesticky Martin,“Application of Neural Networks in Computer Security”, DAAAM International Symposium on Intelligent Manufacturing and Automation, Vol. 69, 1209-1215, 2013.
[6] Navita Agarwal andPrachi Agarwal," Use of Artificial Neural Network in the Field of Security", MIT International Journal of Computer Science & Information Technology", Vol. 3, Issue 1, 42-44, 2013.
[7] Shujuan Jin," Research on Application of Neural Network in Computer Network Security Evaluations",The Open Electrical & Electronic Engineering Journal, 766-771, 2014.
[8] Shakeel PM. Neural Networks Based Prediction Of Wind Energy Using Pitch Angle Control. International Journal of Innovations in Scientific and Engineering Research (IJISER). 2014;1(1):33-7.
[9] S. Jegadeeswari, P. Dinadayalan and N. Gnanambigai," Enhanced Data Security Using Neural Network in Cloud Environment", International Journal of Applied Engineering Research, Vol 11, Issue 1, 278-285, 2016.
[10] J. RubinaParveen, “Neural Networks in Cyber Security”, International Research Journal of Computer Science (IRJCS), Vol 4, Issue 9, 2017.

[11] Sufyan T. Al-Janabi, Belal Al-Khateeb and Ahmed J. Abd", Intelligent Techniques in Cryptanalysis: Review and Future Directions," Original Research Article Uhd Journal of Science and Technology. UHD Journal of Science and Technology, Vol 1, Issue 1, 2017.
[12] MuriloCoutinho, Robson de Oliveira Albuquerque, Fábio Borges, Luis Javier GarcíaVillalba and TaiHoon Kim," learing perfectly secure cryptography to protect communication with Adversarial neural cryptography", MDPI, sensors 2018,18,1306:doi:10.3390\18051306,2018.
[13] Sridhar KP, Baskar S, Shakeel PM, Dhulipala VS., “Developing brain abnormality recognize system using multi-objective pattern producing neural network”, Journal of Ambient Intelligence and Humanized Computing, 2018:1-9.https://doi.org/10.1007/s12652-018-1058-y
[14] Belal Al-Khateeb andMahaMahmood, “A Framework for an Automatic Generation of Neural Networks”, IJCSI International Journal of Computer Science Issues, Vol. 10, Issue 3, No 2, 1694-0784, 2013.
[15] Mohammed, M.A., Al-Khateeb, B., Rashid, A.N., Ibrahim, D.A., Ghani, M.K.A. and Mostafa, S.A., 2018. Neural network and multi-fractal dimension features for breast cancer classification from ultrasound images. Computers & Electrical Engineering.
[16] Obaid,O,I, M.A, Ghani, M.K.A., Mostafa, S.A., &FahadTaha AL-Dhief. “Evaluating the Performance of Machine Learning Techniques in the Classification of Wisconsin Breast Cancer.” International Journal of Engineering & Technology [Online], 7.4.36 (2018): 160-166. Web. 19 Dec. 2018.
[17] Mostafa, S.A., Mustapha, A., Hazeem, A.A., Khaleefah, S.H. and Mohammed, M.A., 2018. An AgentBased Inference Engine for Efficient and Reliable Automated Car Failure Diagnosis Assistance. IEEE Access, 6, pp.8322-8331.
[18] Mostafa, S.A., Ahmad, M.S., Mustapha, A. and Mohammed, M.A., 2017. Formulating layered adjustable autonomy for unmanned aerial vehicles. International Journal of Intelligent Computing and Cybernetics, 10(4), pp.430-450.ORSA Journal on Computing 6(2):154-160.
[19] G. Katz,C.Barrett,D.L.Dill,K.Julianand M.J.Kochenderfer. Reluplex: Anefficientsmt solver for verifying deep neural networks. In International Conference on Computer Aided Verification (CAV), 97-117. Springer, 2017.
[20] A. Meri et al., “Modelling the Utilization of Cloud Health Information Systems in the Iraqi Public Healthcare Sector,” Telemat. Informatics, 2018.
[21] Shakeel, P.M., Tolba, A., Al-Makhadmeh, Zafer Al-Makhadmeh, Mustafa Musa Jaber, “Automatic detection of lung cancer from biomedical data set using discrete AdaBoost optimized ensemble learning generalized neural networks”, Neural Computing and Applications,2019,pp1-14.https://doi.org/10.1007/s00521-018-03972-2
[22] Shakeel PM, Baskar S, Dhulipala VS, Jaber MM., “Cloud based framework for diagnosis of diabetes mellitus using K-means clustering”, Health information science and systems, 2018 Dec 1;6(1):16.https://doi.org/10.1007/s13755-018-0054-0
[23] Shakeel PM, Baskar S, Dhulipala VS, Mishra S, Jaber MM., “Maintaining security and privacy in health care system using learning based Deep-Q-Networks”, Journal of medical systems, 2018 Oct 1;42(10):186.https://doi.org/10.1007/s10916-018-1045-z
[24] M. Jarrar, M. S. Minai, M. Al-Bsheish, A. Meri, and M. Jaber, “Hospital nurse shift length, patientcentered care, and the perceived quality and patient safety,” International Journal of Health Planning and Management, 2018.
[25] Shiqi Wang, Kexin Pei, Justin Whitehouse, Junfeng Yang and Suman Jana, “Formal Security Analysis of Neural Networks using Symbolic Intervals”, Columbia University, 2018.
[26] Khalil Shihab," A Backpropagation Neural Network for Computer Network Security", Journal of Computer Science 2 (9): 710-715, 2006.
[27] A.Manikandan, J.Sakthivel, “Recognizable Proof of Biometric System With Even Distorted And Rectification States”, Journal of Advanced Research in Dynamical and Control Systems, ISSN:1943023X , pp 1393-1398, 2017.

Jour of Adv Research in Dynamical & Control Systems, Vol. 10, 10-Special Issue, 2018