A Lightweight Smart Agent for Detecting Attacks in Distributed Machine Learning

A lightweight intelligent security agent is proposed that functions within the boundaries of a trusted execute environment and controls the detection of anomalies in computing processes when solving object detection tasks in a distributed IT infrastructure. The agent controls the stress factors of the processes and signals their deviations from the standard values using its functionality in the form of an adaptive neuro-fuzzy inference system. The agent's ability to detect rootkits with an accuracy of $96.4 %$ at $3 %$ overhead on processor performance when working with a timeout of 5 seconds is shown. A test bench has been proposed and implemented to study the tasks of trusted distributed machine learning using edge and fog computing.