Home › Glossary › Neural Networks › Backpropagation through time (BPTT)

Intermediate · Neural Networks

Backpropagation through time (BPTT)

Visual diagram · (in preparation) · Math · (in preparation) · Worked example · 3 difficulty levels.

TL;DR. A core algorithm for training recurrent neural networks by unfolding them over time and applying backpropagation.

Technical Definition

A core algorithm for training recurrent neural networks by unfolding them over time and applying backpropagation.

How it works

Backpropagation Through Time (BPTT) is a fundamental algorithm used to train recurrent neural networks (RNNs). It works by 'unrolling' the network into a deep feedforward structure, where each time step is represented, and then applying the standard backpropagation algorithm to compute gradients.

Related Concepts

  • Gradient Descent — An optimization algorithm that iteratively adjusts model parameters by moving in the direction of steepest decrease of the loss function.
  • Backpropagation — An algorithm that efficiently computes gradients by propagating errors backward through the network using the chain rule.
  • Recurrent Neural Network (RNN) — A neural network with loops that maintain hidden state, designed to process sequential data like text and time series.

Further Reading

  • Wikipedia — Glossary of AI