TY - GEN
T1 - MEML
T2 - 12th IEEE/ACM International Conference on Utility and Cloud Computing, UCC Companion 2019
AU - Shalaginov, Andrii
AU - Semeniuta, Oleksandr
AU - Alazab, Mamoun
PY - 2019/12/2
Y1 - 2019/12/2
N2 - Growing number of Smart Applications in recent years bring a completely new landscape of cyber-attacks and exploitation scenario that have not been seen in wild before. Devices in Edge commonly have very limited computational resources and corresponding power source reducing the number of conventional cybersecurity measures available for deployment. This also puts strict requirements on how the signatures of malicious actions can be updated and actualized. It has been proved efficiency of Machine Learning models, Neural Networks in particular, in multiple tasks related to cybersecurity due to the high-abstract precise models and training from historical data. However, when it comes to the devices in Edge, it is clear that the extensive training of the model is not possible, while testing of new unseen data can be successfully done. In addition to the conventional understanding of off-line and on-line model training, this contribution looks into how the Machine Learning can be successfully deployed on IoT while putting unnecessary computations off-chip through parameters transfer over MQTT network, reducing computational footprint on micro-controllers. We believe that proposed approach will be beneficial for many applications in resource-constrained environment.
AB - Growing number of Smart Applications in recent years bring a completely new landscape of cyber-attacks and exploitation scenario that have not been seen in wild before. Devices in Edge commonly have very limited computational resources and corresponding power source reducing the number of conventional cybersecurity measures available for deployment. This also puts strict requirements on how the signatures of malicious actions can be updated and actualized. It has been proved efficiency of Machine Learning models, Neural Networks in particular, in multiple tasks related to cybersecurity due to the high-abstract precise models and training from historical data. However, when it comes to the devices in Edge, it is clear that the extensive training of the model is not possible, while testing of new unseen data can be successfully done. In addition to the conventional understanding of off-line and on-line model training, this contribution looks into how the Machine Learning can be successfully deployed on IoT while putting unnecessary computations off-chip through parameters transfer over MQTT network, reducing computational footprint on micro-controllers. We believe that proposed approach will be beneficial for many applications in resource-constrained environment.
KW - Internet of Things
KW - Machine Learning
KW - Network Security
UR - http://www.scopus.com/inward/record.url?scp=85077303689&partnerID=8YFLogxK
U2 - 10.1145/3368235.3368876
DO - 10.1145/3368235.3368876
M3 - Conference Paper published in Proceedings
SN - 9781450370448
VL - 1
T3 - UCC 2019 Companion - Proceedings of the 12th IEEE/ACM International Conference on Utility and Cloud Computing
SP - 123
EP - 128
BT - UCC 2019 Companion - Proceedings of the 12th IEEE/ACM International Conference on Utility and Cloud Computing
PB - Association for Computing Machinery, Inc
CY - New York
Y2 - 2 December 2019 through 5 December 2019
ER -