The Role of AI in Confronting Cyber(in)Security in loT Ecosystems
No ratings
Presented at ISC2SecureSummit 2019 by
The development and adoption of Internet of Things (IoT) ecosystem have gained enormous traction, as the deployment of sensors, actuators, data, and interface capabilities continue to grow at an alarming rate resulting in a huge amount of generated data. Is it predicted that successful IoT engagements will be an integral part of data analytics and supported by artificial intelligence (AI) capabilities such as machine learning (ML) and deep learning (DL). As the number of interconnected IoT devices is projected to reach billions in the coming years, there have been persistent calls to embrace AI-based algorithms to help secure and manage these devices. Certainly, ML and DL have been an integral part of the IoT cybersecurity landscape to ensure trusted systems for users. The advent of the IoT comes with interconnected sensing devices within the IoT ecosystem that generate data for a wide range of applications and usage. The benefits of current and next generations of IoT cannot be overemphasized. However, the realization of this potential requires data-driven solutions that ensure optimum privacy and security associated with IoT implementations. Despite the value that IoT brings, limitations revolving around privacy and security remain a major concern.Unfortunately, with the advances in intrusion detection methods and technologies, conventional methods for cybersecurity mitigation efforts are not likely to completely prevent the innovative strategies adopted by cybercriminals. The realization of ML and DL in pattern recognition, anomaly detection, and predictive analytics, malware detection, and intrusion detection has impacted how cybersecurity threats and privacy issues pertaining to IoT devices are mitigated. While recent advances in ML algorithms such as linear and logistic regression, decision tree regression, random forest, and support vector machine continue to play an important role in this regard, limitations do exist. The execution of feature extraction of training and test data to produce relevant models can make these algorithms vulnerable to cybersecurity attacks. To address security threats towards machine learning due to potential vulnerabilities, data-driven DL algorithms can work in tandem with IoT for synergy benefits in executing its smart detective, diagnostic, and predictive analytics functions. Among the many AI algorithms, DL has been actively employed in several of these IoT applications. The motivation for this presentation is to discuss DL-based predictive models using TensorFlow to identify, prioritize, and characterize targeted intrusions or attack behaviors to help address privacy and security concerns in IoT ecosystems. TensorFlow is an open source software library known for high-performance numerical computations with support for machine learning and deep learning.