Recurrent Neural Network (RNN)

Models and Architectures

An earlier neural network design for sequences like text and speech - it reads step by step, carrying memory forward.

A recurrent neural network processes sequences one step at a time - word by word, audio frame by audio frame - passing a running memory forward as it goes. That made it the standard choice for language, speech, and time-series data for years.Its weakness was forgetting: over long texts the carried memory fades. Transformers, which look at a whole sequence at once, largely replaced RNNs in language tasks, but the idea still appears in speech processing, forecasting, and lightweight on-device models.