Better Training Data Attribution via Better Inverse Hessian-Vector Products

TL;DR

This paper introduces ASTRA, an efficient algorithm for accurately approximating inverse Hessian-vector products, significantly enhancing training data attribution methods in machine learning.

Contribution

ASTRA employs EKFAC preconditioning with Neumann series to improve iHVP approximation accuracy, requiring fewer iterations and being easier to tune than previous methods.

Findings

ASTRA outperforms existing iHVP approximation methods in accuracy.

Improved iHVP approximation leads to better training data attribution results.

ASTRA is computationally efficient and easier to tune than prior approaches.

Abstract

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.