Pre-training
The first training stage, where a model learns general representations from a large dataset (often unlabelled, via self-supervision) before being adapted to a specific task. Pre-training ideally uses data from the same domain as the eventual task.
Two candidate tools for extracting medication changes from discharge summaries look similar in a demo. One is built on a model pre-trained mostly on general web text, the other on a large body of de-identified clinical notes, so the second has met shorthand like d/c, which can mean discharged or discontinued, far more often and in real notes. Ask each supplier what the pre-training data contained and how recent it is, then test both on a few hundred of your own summaries before comparing prices.
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 →