Active Machine Learning Adversarial Attack Detection in the User Feedback Process

Modern Information and Communication Technology (ICT)-based applications utilize current technological advancements for purposes of streaming data, as a way adapting to the ever-changing landscape. Such efforts require providing accurate, meaningful, trustworthy output from sensors particularly during dynamic virtual sensing. However, ensure that sensing ecosystem is devoid any sensor threats or active attacks, it paramount implement secure real-time strategies. Fundamentally, detection adversarial attacks/instances User Feedback Process (UFP) key forecasting potential attacks in learning. Also, according existing literature, there lacks comprehensive study has focus on an machine learning perspective at time writing this paper. Therefore, authors posit importance detecting strategy. Attack context paper through UFP-Threat driven model been presented action exerts alteration system data. To achieve this, employed ambient data collected smart environment human activity recognition (Continuous Ambient Sensors Dataset, CASA) with fully labeled connections, where we intentionally subject Dataset wrong labels targeted/manipulative attack (by malevolent labeler) UFP, assumption user-labels were connected unique identities. While dataset's classify tasks predict activities, our gives strategies information security point view. Furthermore, modeling have using Meta Language (MAL) compiler detection. The findings experiments conducted shown identification profiling UFP could significantly increase accuracy process high degree certainty paves towards automated approaches Internet Cognitive Things (ICoT).