{"id":9573,"date":"2026-07-27T12:30:27","date_gmt":"2026-07-27T10:30:27","guid":{"rendered":"https:\/\/www.kubicek.ai\/?post_type=lexicon&#038;p=9573"},"modified":"2026-07-27T13:34:03","modified_gmt":"2026-07-27T11:34:03","slug":"lstm-long-short-term-memory","status":"publish","type":"lexicon","link":"https:\/\/www.kubicek.ai\/en\/lexicon\/lstm-long-short-term-memory\/","title":{"rendered":"LSTM (Long Short-Term Memory)"},"content":{"rendered":"<p class=\"wp-block-paragraph\"><strong>LSTM (Long Short-Term Memory)<\/strong> is a type of recurrent neural network designed to overcome the vanishing gradient problem and retain information across long gaps. The key novelty is the separation of two states: alongside the ordinary hidden state, a cell state line runs through the network along which information travels predominantly by addition rather than repeated matrix multiplication. This creates a path on which the gradient survives many time steps. The flow of information is governed by three learned gating mechanisms, each with its own sigmoid. The forget gate decides what portion of the cell&#8217;s existing contents to erase. The input gate determines how much of the newly computed candidate is written into the cell. The output gate selects which part of the contents is passed on as the hidden state at that step. A simplified and cheaper counterpart is the GRU, which merges the gates and states into one. From the 1990s onwards, LSTMs were for roughly two decades the strongest tool for speech, translation and time series, until the transformer architecture displaced them.<\/p>\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n<p class=\"wp-block-paragraph\">Think of a workbench with a conveyor belt carrying a single folder of notes past you. You stand at the belt holding three rulers. With the first you decide which old notes to throw out of the folder. With the second, how much of what you have just overheard to add to it. With the third, what to pull out of the folder and read aloud to a colleague. Because the folder travels on down the belt essentially untouched and you only add or remove the odd thing, the murderer&#8217;s name from chapter two can comfortably reach the final chapter.<\/p>\n","protected":false},"featured_media":0,"template":"","class_list":["post-9573","lexicon","type-lexicon","status-publish","hentry"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/lexicon\/9573","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=9573"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}