Concept · 1 episode(s)

Influence Functions

← all concepts

Definition

Influence functions estimate how a model’s prediction on a given input would change if a specific training point were upweighted or removed. They’re the closest thing we have to attributing model behavior back to the data that caused it — useful, expensive, and still imperfect at scale.

Episodes covering this

Worth reading next

Papers we haven't done a deep dive on yet, but would recommend on this topic.