feedforward

This term is primarily used in technical registers, specifically within control theory and computer science. It describes a proactive approach to system management, contrasting sharply with feedback loops. While feedback reacts to an error after it has occurred, feedforward anticipates the error by measuring the cause of the disturbance and correcting for it before the system output is impacted.

In the context of machine learning, the term describes a specific architecture of neural networks. Here, the "forward" aspect refers to the unidirectional flow of data from the input layer through hidden layers to the output layer, explicitly excluding any loops or cycles that would characterize recurrent networks.