Reservoir computing is a machine-learning approach in which inputs drive a dynamic system, called a reservoir, whose changing internal states help represent patterns in the data. The reservoir may be implemented in software or in physical hardware. Typically, the system itself is not trained in the usual way; instead, a readout layer is trained to map its states to desired outputs.
This approach can make use of complex dynamics for tasks such as processing time-dependent signals. In cyber security, it could be applied to analyzing changing data streams, such as network activity, though its use depends on the system and the task. When discussing circuits, reservoir computing should not be confused with an ordinary resistor-capacitor circuit: a circuit can have evolving electrical states without necessarily performing useful computation. Claims about a particular circuit require evidence that its dynamics support the stated task.