Cognitive artificial intelligence – a complexity based machine learning approach for advanced cyber threats
MetadataShow full item record
Application of machine intelligence is severely challenged in the domain of cyber security due to the surreptitious nature of advanced cyber threats which are persistent and defy existing cyber defense mechanisms. Further, zero day attacks are also on the rise although many of these new attacks are merely a variant of an old and known threat. Machine enabled intelligence is limited in solving advanced and complex problems of detecting these mutated threats. This problem can be attributed to the single scale analysis nature of all the machine learning algorithms including but not limited to artificial neural networks, evolutionary algorithms, bio-inspired machine intelligence et al. This M.Sc. thesis addresses the challenge of detecting advanced cyber threats which conceal themselves under normal or benign activity. Three novel cognitive complexity analysis based algorithms have been proposed which modify the existing single scale machine learning algorithms by incorporating the notion of multiscale complexity in them. Particularly, network based threats are considered using two different publicly available data sets. Moreover, fractal and wavelet based multiscale analysis approach is incorporated in decision making backbone of k-Nearest Neighbours (k-NN) algorithm, Gradient Descent based Artificial Neural Network (ANN), and Hebbian learning algorithm. The classification performance of these algorithms is compared with their traditional single scale counterparts and an improvement in performance is observed consistently. This improvement is attributed to the usage of multiscale based complexity measures in the analysis of algorithm, features and error curve. The notion of multiscale evaluation reveals the hidden relationship which otherwise are averaged out when observed on a single scale. Also, the problem of class overlap which arises due to the stealth nature of cyber-attacks is addressed using the same concept. Conceptually, it is analogous of human cognitive capability employed in pattern discovery from complex objects based on their knowledge about how to connect and correlate various aspects together. It is imperative to note that this multiscale relationship should be a representative of the complexity measure of whole object so that it can characterize patterns based on various scales.
Showing items related by title, author, creator and subject.
Braun, Peter; Cuzzocrea, Alfredo; Keding, Timothy D.; Leung, Carson K.; Padzor, Adam G.M.; Sayson, Dell (Elsevier, 2017)Since its debut in May 2016, Overwatch has quickly become a popular team-based online video game. Despite the popularity of Overwatch, many new players---who join the game unsure how to compete with the game’s veterans---feel ...
Cuzzocrea, Alfredo; MUMOLO, ENZO; FADDA, EDOARDO; Soufargi, Selim; Leung, Carson (CEUR Workshop Proceedings, 2021-09)Computer network systems are often subject to several types of attacks. For example the distributed denial of service (DDoS) attack introduces an excessive traffic load to a web server to make it unusable. A popular method ...
Factors associated with cyber-victimization among immigrants and non-immigrants in Canada: a cross-sectional nationally-representative study Kenny, Kathleen S; Merry, Lisa; Brownbridge, Douglas A; Urquia, Marcelo L (2020-10-16)Abstract Objectives There is a paucity of research on patterns of cyber-victimization in minority groups, including immigrants. This study aimed to identify individual, ...