Influence Functions: Why One Training Point Can Move a Prediction
DEV · 16.11.2026 · 09:23
A model makes a surprising prediction. You want to know why. The usual answers point to features, attention weights, or a saliency map. Those answers explain the input. They do not explain…