What is a recurrent neural network (RNN)?

A Recurrent Neural Network (RNN) is a neural network designed for sequential data, where the order of inputs matters.

Unlike a standard feed-forward network, an RNN keeps a hidden state that carries information from earlier steps in the sequence.

x1 → RNN → h1
          ↓
x2 → RNN → h2
          ↓
x3 → RNN → h3

Here, each hidden state depends on both the current input and the previous hidden state.

Why are RNNs useful?

RNNs are suited to sequences because information from earlier steps can influence later predictions.

They have been used for tasks such as:

• Language modeling
• Speech processing
• Time-series prediction
• Sequence classification

Main limitation

Basic RNNs can struggle with long-range dependencies because gradients may become very small or very large during training.

This is one reason architectures such as LSTM and GRU were developed.

For those variants, see “What is an LSTM network?” and “What is a GRU network?”.

In short, an RNN processes sequence data while carrying information from previous steps through a hidden state.