Neural-network based representation framework for adversary identification in internet of things

Abstrak

Machine learning is one of the potential solutions towards optimizing the security strength towards identifying complex forms of threats in internet of things (IoT). However, a review of existing machine learning-based approaches showcases their sub-optimal performance when exposed to dynamic forms of unseen threats without any a priori information during the training stage. Hence, this manuscript presents a novel machine-learning framework towards potential threat detection capable of identifying the underlying patterns of rapidly evolving threats. The proposed system uses a neural network-based learning model emphasizing representation learning where an explicit masked indexing mechanism is presented for high-level security against unknown and dynamic adversaries. The benchmarked outcome of the study shows to accomplish 11% maximized threat detection accuracy and 33% minimized algorithm processing time.