| dc.description.abstract | The advent of the Internet of Things (IoT), with the consequent changes
in network architectures and communication dynamics, has strongly
conditioned the security market by radically shifting traditional per-
ceptions of the current Internet toward an integrated vision of smart
interconnected objects. However, due to their provided features and
popularity, such devices have become one of the main targets for
attackers who, exploiting several systems-related vulnerabilities and
well-engineered applications, are able to conduct different hostile ac-
tivities. For this reason, also thanks to the great success of Machine
Learning (ML) and Deep Learning (DL) based techniques in the last
decade, many innovative solutions have been proposed in order to
counteract the exponential and yearly growth of malware applica-
tions. However, since the related detection models should provide
an adequate generalization capability, their success strongly depends
on the right choice of the employed features. To this purpose, new
empowered strategies are needed to spot malware threats in several
network security scenarios, with particular attention to those related
to IoT and Federated environments, respectively.
Therefore, this thesis focuses on the enhancement of detection so-
lutions that, due to the presence of many vulnerable and hardware-
constrained devices, are characterized by several challenges regarding
security and privacy. Under this vision, Chapter 1 presents a detailed
overview of the state-of-the-art by highlighting the weaknesses of
the existing approaches. Next, Chapters 2, 3, and 4 focus on the em-
powerment and effectiveness of such solutions by employing new
dynamic and static-based feature representation techniques. Also, they
highlight the capabilities of DL to offer sophisticated models capa-
ble of reducing Run-time damages and involving the computation
capabilities of federated environments, respectively.
On the other hand, due to the recent explosion of IoT-related mal-
ware applications and the necessity of protecting privacy, the thesis
extends its focus to malware detection activities in Federated organi-
zations. Therefore, Chapter 5 proposes a new Markov Chains-based
detector capable of improving the most famous Federated Learning
(FL) based solutions. To this purpose, a dedicated privacy-preserving
architecture is employed, in which the involved clients build the
related detection model by indirectly sharing the analyzed applica-
tions. Finally, Chapter 6 presents the Conclusions about the reported
contributions by highlighting possible and relevant future research
directions. [edited by Author] | it_IT |
