Sequence-to-sequence model
A transformer that uses both encoder and decoder, also called an encoder-decoder model. Built to turn one sequence into another, for tasks like translation, summarisation, and speech recognition. T5 and BART are the well-known examples.
A hospital pilots a sequence-to-sequence model that translates discharge instructions into Spanish for patients who read Spanish more easily than English. The output reads fluently because the decoder writes each word to fit the source text and the words it has already produced, and that fluency can hide a dropped negation or a dose that changed between languages. Before rollout, have qualified medical translators check a few hundred source and translated pairs, including every number, drug name and instruction to stop or avoid something. Most US hospitals also fall under a federal rule that machine translation of text where accuracy is essential must be reviewed by a qualified human translator.
Terms like this come up in real clinical scenarios across the HelloAI courses: bite-sized modules with verifiable certificates. An account takes one minute, no password needed.
Sign in →