Learning for cybersecurity in IoT
Published : 19 December 2016
The goal of this project is to propose a lightweight Security Audit Module capable of making low-cost, pervasive IoT nodes tolerant to attacks. Targeted IoT nodes consist of a processor, a radio interface, and a set of sensors/actuators. The module will identify physical and logical attacks on the node and it will react accordingly, given a desired security level.
We will employ machine learning techniques to model normal and attack behavior starting with a set of hardware and software probes. To realise this, three steps are required:
The first is the definition and categorisation of the attack scenarios and security reactions that will be covered, given the specifications of an IoT node.
The second is the investigation of the software and hardware probes that can be used to determine attack cases.
The final step consists in studying low-cost classifiers for on-line attack detection, and evaluating their performance in terms of
e.g., false positive and false negative, algorithm complexity and overhead.