Showing cs.LGShow all
2 papers · 1 filter
cs.LG2026
Influence Is the Target-Response Jacobian Times Residuals
Paul Geertsema, Simon Geertsema
We introduce the target-response Jacobian , which measures how model predictions respond to changes in individual training targets. comple…
cs.LG2023
AXIL: Exact Instance Attribution for Gradient Boosting
Paul Geertsema, Helen Lu
We derive an exact, prediction-specific instance-attribution method for fitted gradient boosting machines (GBMs) trained with squared-error loss, with the learned tree structure he…