Backpropagation
The core algorithm used to train neural networks by calculating how much each connection contributed to an error and adjusting it accordingly.
Explanation
Backpropagation, short for backward propagation of errors, is the method neural networks use to learn: after the network makes a prediction, the error is measured and passed backward through the layers of the network, and each connection's weight is nudged to reduce future error. The method was independently discovered by several researchers, but the influential 1986 paper 'Learning representations by back-propagating errors' by David Rumelhart, Geoffrey Hinton, and Ronald Williams is what brought it to wide attention and made it standard practice in neural network research. Marketers working with AI rarely need the term itself, but backpropagation is the training mechanism behind the machine learning models, recommendation engines, ad targeting, generative AI, that now sit underneath most modern marketing technology.
Usage context / examples
A recommendation engine that improves its personalized product suggestions over time is doing so because backpropagation is continually adjusting the underlying model based on which past recommendations led to clicks or purchases.
Related terms
Neural network, machine learning, gradient descent, deep learning
Turkish equivalent
Geri yayılım
Turkish definition
Backpropagation, bir yapay sinir ağının hatasını geriye doğru yayarak her bağlantının ağırlığını düzeltmesini sağlayan öğrenme algoritmasıdır.
Source: MMA Academy Sözlük