Based on the least-square minimization a computationally efficient learning algorithm for the Principal Component Analysis(PCA) is derived. The dual learning rate parameters are adaptively introduced to make the propo...Based on the least-square minimization a computationally efficient learning algorithm for the Principal Component Analysis(PCA) is derived. The dual learning rate parameters are adaptively introduced to make the proposed algorithm providing the capability of the fast convergence and high accuracy for extracting all the principal components. It is shown that all the information needed for PCA can be completely represented by the unnormalized weight vector which is updated based only on the corresponding neuron input-output product. The convergence performance of the proposed algorithm is briefly analyzed.The relation between Oja’s rule and the least squares learning rule is also established. Finally, a simulation example is given to illustrate the effectiveness of this algorithm for PCA.展开更多
This article discusses vision recognition process and finds out that human recognizes objects not by their isolated features, but by their main difference features which people get by contrasting them. According to th...This article discusses vision recognition process and finds out that human recognizes objects not by their isolated features, but by their main difference features which people get by contrasting them. According to the resolving character of difference features for vision recognition, the difference feature neural network(DFNN) which is the improved auto-associative neural network is proposed.Using ORL database, the comparative experiment for face recognition with face images and the ones added Gaussian noise is performed, and the result shows that DFNN is better than the auto-associative neural network and it proves DFNN is more efficient.展开更多
Purpose–The task of internet intrusion detection is to detect anomalous network connections caused by intrusive activities.There have been many intrusion detection schemes proposed,most of which apply both normal and...Purpose–The task of internet intrusion detection is to detect anomalous network connections caused by intrusive activities.There have been many intrusion detection schemes proposed,most of which apply both normal and intrusion data to construct classifiers.However,normal data and intrusion data are often seriously imbalanced because intrusive connection data are usually difficult to collect.Internet intrusion detection can be considered as a novelty detection problem,which is the identification of new or unknown data,to which a learning system has not been exposed during training.This paper aims to address this issue.Design/methodology/approach–In this paper,a novelty detection-based intrusion detection system is proposed by combining the self-organizing map(SOM)and the kernel auto-associator(KAA)model proposed earlier by the first author.The KAA model is a generalization of auto-associative networks by training to recall the inputs through kernel subspace.For anomaly detection,the SOM organizes the prototypes of samples while the KAA provides data description for the normal connection patterns.The hybrid SOM/KAA model can also be applied to classify different types of attacks.Findings–Using the KDD CUP,1999 dataset,the performance of the proposed scheme in separating normal connection patterns from intrusive connection patterns was compared with some state-of-art novelty detection methods,showing marked improvements in terms of the high intrusion detection accuracy and low false positives.Simulations on the classification of attack categories also demonstrate favorable results of the accuracy,which are comparable to the entries from the KDD CUP,1999 data mining competition.Originality/value–The hybrid model of SOM and the KAA model can achieve significant results for intrusion detection.展开更多
基金Supported by the National Natural Science Foundation of Chinathe Science foundation of Guangxi Educational Administration
文摘Based on the least-square minimization a computationally efficient learning algorithm for the Principal Component Analysis(PCA) is derived. The dual learning rate parameters are adaptively introduced to make the proposed algorithm providing the capability of the fast convergence and high accuracy for extracting all the principal components. It is shown that all the information needed for PCA can be completely represented by the unnormalized weight vector which is updated based only on the corresponding neuron input-output product. The convergence performance of the proposed algorithm is briefly analyzed.The relation between Oja’s rule and the least squares learning rule is also established. Finally, a simulation example is given to illustrate the effectiveness of this algorithm for PCA.
文摘This article discusses vision recognition process and finds out that human recognizes objects not by their isolated features, but by their main difference features which people get by contrasting them. According to the resolving character of difference features for vision recognition, the difference feature neural network(DFNN) which is the improved auto-associative neural network is proposed.Using ORL database, the comparative experiment for face recognition with face images and the ones added Gaussian noise is performed, and the result shows that DFNN is better than the auto-associative neural network and it proves DFNN is more efficient.
基金Suzhou Municipal Science and Technology Foundation Key Technologies for Video Objects Intelligent Analysis for Criminal Investigation(SS201109).
文摘Purpose–The task of internet intrusion detection is to detect anomalous network connections caused by intrusive activities.There have been many intrusion detection schemes proposed,most of which apply both normal and intrusion data to construct classifiers.However,normal data and intrusion data are often seriously imbalanced because intrusive connection data are usually difficult to collect.Internet intrusion detection can be considered as a novelty detection problem,which is the identification of new or unknown data,to which a learning system has not been exposed during training.This paper aims to address this issue.Design/methodology/approach–In this paper,a novelty detection-based intrusion detection system is proposed by combining the self-organizing map(SOM)and the kernel auto-associator(KAA)model proposed earlier by the first author.The KAA model is a generalization of auto-associative networks by training to recall the inputs through kernel subspace.For anomaly detection,the SOM organizes the prototypes of samples while the KAA provides data description for the normal connection patterns.The hybrid SOM/KAA model can also be applied to classify different types of attacks.Findings–Using the KDD CUP,1999 dataset,the performance of the proposed scheme in separating normal connection patterns from intrusive connection patterns was compared with some state-of-art novelty detection methods,showing marked improvements in terms of the high intrusion detection accuracy and low false positives.Simulations on the classification of attack categories also demonstrate favorable results of the accuracy,which are comparable to the entries from the KDD CUP,1999 data mining competition.Originality/value–The hybrid model of SOM and the KAA model can achieve significant results for intrusion detection.