Self-attention
Attention where the query, key, and value all come from the same input, so a sequence attends to itself. It is how a transformer works out the contextual relationships among the words within one sequence.
A coding assistant suggests one set of diagnosis codes for a progress note sent alone and a slightly different set when the same note arrives with a template header and yesterday's note attached. In self-attention each token's representation is computed from other tokens in the same input, so text that looks irrelevant can still shift how the important sentences are read. Validate it with inputs assembled exactly as production will send them, templates and history included, and retest whenever the interface changes what it sends.
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 →