Better Training Data Attribution via Better Inverse Hessian-Vector Products

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Wang, Andrew, Nguyen, Elisa, Yang, Runshi, Bae, Juhan, McIlraith, Sheila A., Grosse, Roger
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866912492812238848
author Wang, Andrew
Nguyen, Elisa
Yang, Runshi
Bae, Juhan
McIlraith, Sheila A.
Grosse, Roger
author_facet Wang, Andrew
Nguyen, Elisa
Yang, Runshi
Bae, Juhan
McIlraith, Sheila A.
Grosse, Roger
contents Training data attribution (TDA) provides insights into which training data is responsible for a learned model behavior. Gradient-based TDA methods such as influence functions and unrolled differentiation both involve a computation that resembles an inverse Hessian-vector product (iHVP), which is difficult to approximate efficiently. We introduce an algorithm (ASTRA) which uses the EKFAC-preconditioner on Neumann series iterations to arrive at an accurate iHVP approximation for TDA. ASTRA is easy to tune, requires fewer iterations than Neumann series iterations, and is more accurate than EKFAC-based approximations. Using ASTRA, we show that improving the accuracy of the iHVP approximation can significantly improve TDA performance.
format Preprint
id arxiv_https___arxiv_org_abs_2507_14740
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Better Training Data Attribution via Better Inverse Hessian-Vector Products
Wang, Andrew
Nguyen, Elisa
Yang, Runshi
Bae, Juhan
McIlraith, Sheila A.
Grosse, Roger
Machine Learning
Training data attribution (TDA) provides insights into which training data is responsible for a learned model behavior. Gradient-based TDA methods such as influence functions and unrolled differentiation both involve a computation that resembles an inverse Hessian-vector product (iHVP), which is difficult to approximate efficiently. We introduce an algorithm (ASTRA) which uses the EKFAC-preconditioner on Neumann series iterations to arrive at an accurate iHVP approximation for TDA. ASTRA is easy to tune, requires fewer iterations than Neumann series iterations, and is more accurate than EKFAC-based approximations. Using ASTRA, we show that improving the accuracy of the iHVP approximation can significantly improve TDA performance.
title Better Training Data Attribution via Better Inverse Hessian-Vector Products
topic Machine Learning
url https://arxiv.org/abs/2507.14740