Deep learning (DL) has been used globally in almost every sector of technology and society. Despite its huge success, DL models and applications have been susceptible to adversarial attacks, impacting the accuracy and integrity of these models. Many state-of-the-art models are vulnerable to attacks by well-crafted adversarial examples, which are perturbed versions of clean data with a small amount of noise added, imperceptible to the human eyes, and can quite easily fool the targeted model. This paper introduces six most effective gradient-based adversarial attacks on the ResNet image recognition model, and demonstrates the limitations of traditional adversarial retraining technique. The authors then present a novel ensemble defense strategy based on adversarial retraining technique. The proposed method is capable of withstanding the six adversarial attacks on cifar10 dataset with accuracy greater than 89.31% and as high as 96.24%. The authors believe the design methodologies and experiments demonstrated are widely applicable to other domains of machine learning, DL, and computation intelligence securities.
1 April 2021 in International Journal of Cloud Applications and Computing
Social networks allow people to connect to one another. Over time, these accounts become an essential part of one's online identity. The account stores various personal data and contains one's network of acquaintances. Attackers seek to compromise user accounts for various malicious purposes, such as distributing spam, phishing, and much more. Timely detection of compromises becomes crucial for protecting users and social networks. This article proposes a novel system for detecting compromises of a social network account by considering both post behavior and textual content. A deep multi-layer perceptron-based autoencoder is leveraged to consolidate diverse features and extract underlying relationships. Experiments show that the proposed system outperforms previous techniques that considered only behavioral information. The authors believe that this work is well-timed, significant especially in the world that has been largely locked down by the COVID-19 pandemic and thus depends much more on reliable social networks to stay connected.
Wireless sensor network applications in healthcare and biomedical technology have received increasing attention, while associated security and privacy issues remain open areas of consideration. The relevance of this technology to our growing elderly population, as well as our increasingly over-crowded and attention-drained healthcare systems, is promising. However, prior to the emergence of these systems as a ubiquitous technology, healthcare providers and regulatory agencies must determine an acceptable level of security and privacy. This paper will review biomedical applications of wireless sensor networks, identify security and privacy issues to be addressed, and note some of the proposed methods for securing these systems.
1 January 2006 in International Journal of Sensor Networks
One major challenge of contention-based Medium-Access Control (MAC) for sensor networks is to reduce energy waste in idle listening time without global synchronisation of active/sleep duty cycles. This paper presents Distributed Power Scheduling (DPS), a medium access control protocol that is specifically designed for a sensor network that supports data aggregation. DPS integrates data aggregation process into power-mode scheduling in the MAC layer and effectively reduces packet delay. Two enhancements, adaptive aggregation and adaptive sleeping, dynamically adjust aggregation period and sleeping time, and greatly reduce idle listening period. DPS requires only local synchronisation and supports self-configuration. Detailed protocol description is presented. Formal analysis presents upper bounds on expected frame time (duty cycle period)and expected packet delay. Simulation study shows that the two enhancements in DPS achieved 28% reduction in idle listening time, which resulted in 20% energy reduction, as compared to the basic DPS protocol.
18 September 2017 in Network Modeling Analysis in Health Informatics and Bioinformatics