Recurrent Neural Network (RNN)

A recurrent neural network, or RNN, processes sequential data while carrying information from earlier steps into later calculations. This internal state allows it to model order and context in text, speech, sensor streams, and time series. Basic RNNs can struggle to preserve information across long sequences because gradients may vanish or explode during training. LSTM and GRU architectures introduced gates that help control what is remembered or forgotten. Transformers now dominate many language tasks because they process long-range relationships more efficiently in parallel, but recurrent models remain useful where streaming, compact state, or step-by-step processing matters. Architecture choice depends on sequence length, latency, hardware, data volume, and required accuracy.