Revisit, Extend, and Enhance Hessian-Free Influence Functions
This work addresses the challenge of applying influence functions to deep models by improving a simple approximation method, offering incremental advancements for researchers and practitioners in model interpretation and robustness.
The paper revisits TracIn, a naive approximation method for influence functions that replaces the inverse Hessian with an identity matrix, providing insights into its effectiveness and extending it to fairness and robustness applications with enhancements via ensemble strategies, validated through experiments on synthetic data, noisy label detection, sample selection for LLM fine-tuning, and adversarial defense.
Influence functions serve as crucial tools for assessing sample influence in model interpretation, subset training set selection, noisy label detection, and more. By employing the first-order Taylor extension, influence functions can estimate sample influence without the need for expensive model retraining. However, applying influence functions directly to deep models presents challenges, primarily due to the non-convex nature of the loss function and the large size of model parameters. This difficulty not only makes computing the inverse of the Hessian matrix costly but also renders it non-existent in some cases. Various approaches, including matrix decomposition, have been explored to expedite and approximate the inversion of the Hessian matrix, with the aim of making influence functions applicable to deep models. In this paper, we revisit a specific, albeit naive, yet effective approximation method known as TracIn. This method substitutes the inverse of the Hessian matrix with an identity matrix. We provide deeper insights into why this simple approximation method performs well. Furthermore, we extend its applications beyond measuring model utility to include considerations of fairness and robustness. Finally, we enhance TracIn through an ensemble strategy. To validate its effectiveness, we conduct experiments on synthetic data and extensive evaluations on noisy label detection, sample selection for large language model fine-tuning, and defense against adversarial attacks.