A fundamental learning algorithm used to train artificial neural networks. During training, backpropagation computes how adjusting each network weight would change the output error, and then updates weights via gradient descent. In forex robots using deep learning, backpropagation is how the model “learns” from price data: the bot compares its predicted output to the actual result, then propagates the error backward through the network to adjust weights for next time.

Back to Robots Glossary Index