Slide-level

Predict a slide- or patient-level label from a bag of tile features. You build the slide representation one of two ways, then attach a task head. The walkthrough runs both paths end to end and shows that classification, regression, and survival are just a different head on the same extracted features.

Method	Summary	Walkthrough
Tile encoder + MIL aggregator	A tile-level encoder emits one vector per tile; an aggregator (ABMIL, CLAM, TransMIL, …) pools the bag into a slide vector before the head.	Slide-level
Slide-level encoder	A slide-native encoder emits one vector per slide directly — no aggregator; the task head consumes it as-is.	Slide-level

See also

Reference docs — task heads and metrics on Tasks, Classification, Regression, and Survival; the aggregator zoo on Aggregators.