About this document

Better Hessians for Influence Functions by Ten Nguyen Van is a document available to read on EtoBox.

This dissertation explores the trade-offs in curvature approximation for influence functions in machine learning, particularly addressing the challenges posed by the Hessian bottleneck in deep learning models. It investigates how different Hessian approximations affect the quality of influence-function attributions, which are crucial for understanding model behavior and debugging. The work includes a mathematical framework, computational challenges, and experimental evaluations to determine the fidelity of

Author

Ten Nguyen Van

Language

EN