Inference
What happens when a trained model is actually used, taking an input and producing an output. It is the counterpart to training: training is how the model is built, inference is every time it is run afterward. The word comes up around cost and speed, because each use consumes computing power, and a tool that is cheap to use once can be expensive across a whole health system. For clinical use the practical point is that the model is not learning from your case during inference unless it has been specifically set up to, so it is applying what it already learned, not adapting to you in the moment.
An ambient documentation tool is piloted with five clinicians, and the inference bill is small enough to sit in one department's budget. Each note is a separate run of the model, and if the contract charges per note, then at an illustrative 50 cents each, 500 notes a week costs $250, while 40,000 notes a week across all outpatient clinics costs $20,000, just over a million dollars a year. Ask for pricing at your full expected volume, build the business case on that figure, and check how response times hold up when every clinic is using it on a busy Monday morning.
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 →