Metadata (cookie, IP address, location) may be collected for the functioning of the site. If you do not want this data processed, then you should leave the site.
OK
Artificial immune networks as a basis for artificial intelligence
The most important property for artificial intelligence (AI) is the ability to quickly and robustly learn from a small number of examples, which means the ability of AI to process large volumes of data as well as form reliable decisions and make highly accurate predictions, even if the training sample is limited in size and not fully representative.
The project proposes a bioinspired approach as an alternative to neural networks, which can be used with small volumes and low representativeness of training samples.
The project proposes a bioinspired approach as an alternative to neural networks, which can be used with small volumes and low representativeness of training samples.
new model of AIN
in machine learning
An abstract model of an artificial immune network (AIN) based on a committee of classifiers and stable algorithms for its learning (supervised and reinforcement) for classification problems is proposed.
The developed AIN model has emergence, memory, double plasticity, and robust learning. Robust refers to a low tendency to overfit
The developed AIN model has emergence, memory, double plasticity, and robust learning. Robust refers to a low tendency to overfit
AIN in biometrics
and other tasks
"Deep" neural networks require large amounts of training samples. But there are many problems characterized by small amounts of experimental data. For example, sampling for medical research involves the need to verify the patient’s disease, which is often associated with invasive research. Samples of sufficient size are usually collected over many years.
The problem of a small volume and low representativeness of training samples most clearly manifests itself when creating biometric identification and authentication machines. The specificity of these tasks is that setting up a biometric system must be done quickly (the user cannot be required to repeat data entry many times; otherwise, the system will not be used in practice). This example is typical in that the problem of lack of sampling will not disappear in the future, regardless of the amount of biometric data accumulated by researchers around the world. In real practice, the system will still be trained on a small number of examples (from 10 to 30).
The problem of a small volume and low representativeness of training samples most clearly manifests itself when creating biometric identification and authentication machines. The specificity of these tasks is that setting up a biometric system must be done quickly (the user cannot be required to repeat data entry many times; otherwise, the system will not be used in practice). This example is typical in that the problem of lack of sampling will not disappear in the future, regardless of the amount of biometric data accumulated by researchers around the world. In real practice, the system will still be trained on a small number of examples (from 10 to 30).
Key publications
on the subject "Artificial immune networks"
An abstract model of an artificial immune network based on a classifier committee for biometric pattern recognition by the example of keystroke dynamics
biometric authentication, bagging, boosting, feature subspaces, machine learning on small samples, ensembles of models.
http://www.mathnet.ru/links/e12035696decffaaf458449f2cfe2a48/co853.pdf
http://www.mathnet.ru/links/e12035696decffaaf458449f2cfe2a48/co853.pdf
Sulavko AE. An abstract model of an artificial immune network based on a classifiers committee for biometric pattern recognition by the example of keystroke dynamics. Computer Optics 2020; 44(5): 830-842. DOI: 10.18287/2412-6179-CO-717.
An abstract model of an artificial immune network (AIS) based on a classifier committee and robust learning algorithms (with and without a teacher) for classification problems, which are characterized by small volumes and low representativeness of training samples, are proposed. Evaluation of the effectiveness of the model and algorithms is carried out by the example of the authentication task using keyboard handwriting using 3 databases of biometric metrics. The AIS developed possesses emergence, memory, double plasticity, and stability of learning. Experiments have shown that AIS gives a smaller or comparable percentage of errors with a much smaller training sample than neural networks with certain architectures.
Bayes-Minkowski measure and building on its basis immune machine learning algorithms for biometric facial identification
pattern recognition, feature space, biometrics, correlation dependence of features, informativeness of features, robustness of machine learning
https://iopscience.iop.org/article/10.1088/1742-6596/1546/1/012103/pdf
https://iopscience.iop.org/article/10.1088/1742-6596/1546/1/012103/pdf
Sulavko A.E. Bayes-Minkowski measure and building on its basis immune machine learning algorithms for biometric facial identification // Journal of Physics: Conference Series. - Vol. 1546. - IV International Scientific and Technical Conference "Mechanical Science and Technology Update" (MSTU-2020) 17-19 March, 2020, Omsk, Russian Federation. - doi:10.1088/1742-6596/1546/1/012103
In this paper we propose a new Bayes-Minkowski proximity measure that can be used to process correlated biometric, biomedical, and other type of data (with the normal distribution law or close to it). The Bayes-Minkowski measure is an antagonist criterion with respect to the Minkowski measure, since it shows opposite properties. It is possible to build a hybrid network of classifiers and apply immune learning algorithms to the network based on these proximity measures. It was demonstrated in the work on the example of tasks of identification and verification of a person’s personality by facial image. The achieved errors probabilities of person’s identification and verification by face features were: 0029 и 0.0017, respectively.
Artificial intelligence in a protected design based on immune network models of pattern recognition using the example of biometrics-code converters
keyboard handwriting, signature, facial image, neural networks, classifier committees, machine learning, secure neural network containers
view the work on researchgate.net
view the work on researchgate.net
Shalina E.V., Malinin N.V., Sulavko A.E., Stadnikov D.G. Artificial intelligence in a secure design based on immune network models of pattern recognition using the example of biometrics-code converters // Issues of information protection. – 2020. - No. 2. – pp. 31-40
It is shown that, based on the models of artificial immune networks proposed by the authors, it is potentially possible to build artificial intelligence (AI) systems that are resistant to attempts to perform the following actions by any unauthorized person: analysis of operations performed by AI, unauthorized control of AI, extraction of knowledge from AI. AI tasks related to biometric identification and authentication are considered.
Intelligent network attack detection system based on artificial immune system mechanisms
attack detection system, artificial immune system, KDD99, information security, network security, network attack
https://moit.vivt.ru/?p=8637&lang=ru
https://moit.vivt.ru/?p=8637&lang=ru
V.I. Vasiliev, R.R. Shamsutdinov. Intelligent system for detecting network attacks based on artificial immune system mechanisms // Modeling, optimization and information technologies. – 2019. - No. 1. – pp. 521-535
The article is devoted to the problem of detecting network attacks, both known and previously unknown. The application of various methods of artificial intelligence to solve this problem in the scientific literature has been analyzed, the advantages of the artificial immune system have been identified, and its main mechanisms have been analyzed: generation and negative selection of artificial lymphocytes, their periodic renewal, determination of the fact of their response, and clonal selection of reacting lymphocytes; describes a developed attack detection system based on an artificial immune system, containing a sniffing subsystem, which allows it to analyze real data about network connections at the host level. The KDD99 network connection data set was also described, using which the effectiveness of the developed system was assessed. The scientific literature offering methods for compressing the original data set was analyzed, the shortcomings of the proposed methods were identified, and an independent experimental determination of the significant parameters of network connections contained in the data set was carried out. 13 significant parameters out of 41 were identified. After the preliminary processing and preparation of the analyzed data, a series of experiments were described, the results of which determined the high efficiency of the developed system in detecting unknown network attacks and classifying known attacks.
Immune algorithms for pattern recognition and their application in biometric systems (Review)
pattern recognition, biometrics, feature space, learning, immune networks, dendritic cells, negative selection, clonal selection
view the work on researchgate.net
view the work on researchgate.net
Shalina E.V., Malinin N.V., Sulavko A.E., Stadnikov D.G. Artificial intelligence in a secure design based on immune network models of pattern recognition using the example of biometrics-code converters // Issues of information protection. – 2020. - No. 2. – pp. 31-40
Existing algorithms used for data mining and pattern recognition, based on the principles of the immune system, are summarized and described. The main results of the use of immune algorithms for solving problems of biometric identification and authentication are summarized. The advantages of artificial immune systems compared to artificial neural networks and “fuzzy extractors”, which are used to create biometric authentication systems, are described. The problem of safe storage of biometric standards when using immune algorithms and options for its possible solution are considered.
