You are cordially invited to the following research seminar by Mert Nakıp.
Title: Online Self-Supervised Learning Intrusion Detection Towards Secure Internet of Things
Time: 8 Nov. 2023, 10:30
Place: EEE Seminar Room (D-51)
Summary: Secure Internet of Things (IoT) systems are extremely difficult to achieve as most IoT devices are low-cost and low-maintenance devices with low computing power and human intervention to run complex security methods. Therefore, lightweight data-driven, especially Machine Learning (ML)-based, security methods have been developed particularly for IoT systems. On the other hand, these methods often learn offline from large data collected through extensive simulations, which may be time consuming and provide biased (misleading) data. This presentation outlines the research on the open issues and ways to enable fully online and lightweight learning for ML-based intrusion detection paving the way towards secure IoT. This research first develops an Intrusion Detection System (IDS) with offline and quasi-online learning, and it evaluates the performance of this IDS for Botnet DDoS, DoS, and zero-day attacks. This IDS learns the normal traffic patterns of the IoT network and detects both malicious network traffic packets and compromised IoT devices. Then, the novel Self-Supervised Intrusion Detection (SSID) framework is proposed to enable fully online learning of ML-based IDS requiring no human intervention. The SSID framework collects and labels traffic packets based only on the decisions of the IDS and their statistically measured trustworthiness. The SSID framework enables IDS to adapt time-varying characteristics of the network traffic quickly, eliminates the need for offline data collection, prevents human errors in data labeling, and avoids labor costs for model training and data collection through experiments. Therefore – as the experimental results on public datasets for malicious traffic and compromised device detection also suggest – SSID is very useful and advantageous to develop an online learning ML-based IDS for IoT systems.
Short-Bio: I obtained my B.Sc. degree, graduating ranked first in my class from the Electrical-Electronics Engineering at Yaşar University (Izmir, Türkiye) in 2018. My design of a multisensor fire detector via machine learning methods was ranked #1 nationally at the Industry-Focused Undergraduate Graduation Projects Competition organized by TÜBİTAK. I completed my M.Sc. thesis in Electrical-Electronics Engineering at Yaşar University (Izmir, Türkiye) in 2020. My thesis focused on the application of machine learning methods to IoT and was supported by the National Graduate Scholarship Program of TÜBİTAK 2210C in High-Priority Technological Areas. Since 2020, I am a Research Assistant and Ph.D. Candidate at the Institute of Theoretical and Applied Informatics, Polish Academy of Sciences (Gliwice, Poland). I am currently a researcher in the European research and innovation project, named DOSS. I was also a researcher in the IoTAC Research and Innovation Action of the European Commission H2020 Program between 2020 and 2023.