Encoder
The part of a transformer that reads an input sequence and turns each word into a contextual representation, a vector that reflects the rest of the input. It stacks identical layers, six in the original transformer and 12 in BERT-base, each combining multi-head self-attention with a feed-forward network. When one summary of the whole input is needed, for example to classify a report, it is drawn from those outputs, usually from a special classification token or an average of all of them.
A vendor explains that its chest CT products for nodules, emphysema and coronary calcium all run on one shared image encoder, pretrained once on a large collection of scans, with a small task-specific layer on top of it for each product. A scanner type, reconstruction kernel or patient group that was thin in the encoder's pretraining data can then weaken every product built on it at once. Ask what data the shared encoder saw, and validate each product locally, since one good result says little about the others.
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 →