Evolutionary optimization of on-line multilayer perceptron for similarity-based access control

Abstract

Neural Networks have been successfully used in different fields of Information Security such that network intrusion detection and malware analysis because of ability to provide high level of abstraction for complex and incomplete data. Despite its successful application as off-line learning method, the on-line learning can be challenging when dealing with data streams. This paper presents an ongoing research on on-line Neural Network for Access Control. It can be used for similarity-based access to sensitive information. Conventional training is not efficient when dealing with data streams such that access patterns flow since the availability of the data samples is limited. Considering this obstacle we proposed to use Genetic Algorithm as meta-heuristic optimization in selection of individual training rates α for each weight. Similarity-based Access Control mechanism deals with a data stream that includes continuous flow of attributes characterizing user and resources, so the task is to estimate the likelihood of legitimacy of user accessing a particular resource in dynamic environment. This research contributes to the field of Information Security by overcoming the limitations of data stream mining in agile environment.