Next Generation Automated Reservoir Computing for Cyber Defense

It is a fact, that important information related to systems' behavior and dynamics, can be revealed as time passes. Observing changes over time, can often lead to the detection of patterns and trends that might not be immediately apparent from a single system's snapshot. Additionally, the concept of time can be essential in understanding cause-and-effect relationships. Observing how changes in one variable over time affect changes in another, can gain insights into the causal relationships between different system components. Time series analysis of data that change over time, can be a powerful tool for understanding complex systems in the field of cyber security. Reservoir Computing (RC) is a Machine Learning technique, using a fixed and randomly generated high-dimensional dynamic system, called a Reservoir, to transform and classify input data. The reservoir acts as a nonlinear and temporal filter of the input data, which is then readout by a linear output layer. Continuous-Time Reservoir Computing (CTRC) is a type of recurrent neural network, aiming to model the continuous-time dynamics of the network's neurons. It is particularly useful for applications where time is critical and it can provide insights into the underlying system's dynamics. This paper proposes a next-generation CTRC for cyber defense, where the reservoir neurons are modeled as continuous-time dynamical systems. This means that their behavior is described by a system of differential equations that change over time. In order to model the drift phenomenon, identify the abnormal changes in the data, and adaptively stabilize the learning system. The CTRC parameters are optimized using the Reinforcement Learning (RL) method. The proposed system, as proved experimentally, has several advantages over discrete systems, including the ability to handle signals with high sampling rates and to effectively capture real cyber security signals' continuous nature.