{"id":9610,"date":"2025-07-10T12:30:33","date_gmt":"2025-07-10T10:30:33","guid":{"rendered":"https:\/\/www.kubicek.ai\/?post_type=lexicon&#038;p=9610"},"modified":"2026-07-27T13:50:49","modified_gmt":"2026-07-27T11:50:49","slug":"backpropagation","status":"publish","type":"lexicon","link":"https:\/\/www.kubicek.ai\/en\/lexicon\/backpropagation\/","title":{"rendered":"Backpropagation"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\"><strong>Backpropagation<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Backpropagation is the key algorithm by which neural networks learn. After the network performs a task \u2013 recognising an image, say \u2013 it compares its result with the correct answer and computes how far it was wrong. It then sends this error information back through its whole structure, from the end to the beginning. On this backward pass it determines which internal connections (weights) bear the greatest responsibility for the error and adjusts them slightly so that the next attempt turns out better.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">You can picture it simply as a game of darts. When you throw for the first time and miss the target, you look at where the dart landed (you measure the error). Then you think about what you did wrong \u2013 perhaps your stance was off or you released the dart too early (you analyse what contributed to the error). For the next throw you learn from this and adjust your style slightly (you adjust the weights). You repeat the process until you start hitting the bullseye.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Backpropagation<\/strong> <em>(backward propagation of errors)<\/em><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Technical definition:<\/strong> Backpropagation is an optimisation algorithm used to train artificial neural networks, especially in the context of supervised learning. It is an efficient method for computing the gradient of the error function with respect to all weights and biases in the network. The algorithm has two main phases. In the first, the <em>forward pass<\/em>, the input data are processed by the network and an output is generated. In the second, the <em>backward pass<\/em>, the error (the difference between predicted and actual output) is propagated from the output layer back towards the input layers. Using the chain rule from differential calculus, the partial derivatives of the error with respect to individual weights are computed step by step, determining their influence on the total error. This computed gradient is then used by an optimisation method such as gradient descent to update the weights so as to minimise the network&#8217;s error.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Etymology and context:<\/strong> The name precisely describes the core of the method: propagating information about the error backwards, that is, against the direction in which the signal travelled during computation. The basic principles were laid down as early as the 1960s in the context of control theory and dynamic programming. The algorithm as we know it today was described and popularised for neural networks in 1986 by David Rumelhart, Geoffrey Hinton and Ronald Williams. Their work showed that backpropagation allows deep networks to learn useful internal representations of data, which became the foundation of the deep learning revolution.<\/p>\n","protected":false},"featured_media":0,"template":"","class_list":["post-9610","lexicon","type-lexicon","status-publish","hentry"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/lexicon\/9610","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/lexicon"}],"about":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/types\/lexicon"}],"wp:attachment":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/media?parent=9610"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}