Final-Model-Only Data Attribution with a Unifying View of Gradient-Based Methods
This addresses the challenge of data attribution in machine learning for practitioners with limited access to training details, but it is incremental as it builds on existing gradient-based methods.
The paper tackles the problem of attributing model behavior to training data when only the final trained model is available, proposing further training as a gold standard and unifying existing gradient-based methods as approximations to it. They found that first-order methods' approximation quality decays with more training, while influence function methods are more stable but lower in quality.
Training data attribution (TDA) is the task of attributing model behavior to elements in the training data. This paper draws attention to the common setting where one has access only to the final trained model, and not the training algorithm or intermediate information from training. To serve as a gold standard for TDA in this "final-model-only" setting, we propose further training, with appropriate adjustment and averaging, to measure the sensitivity of the given model to training instances. We then unify existing gradient-based methods for TDA by showing that they all approximate the further training gold standard in different ways. We investigate empirically the quality of these gradient-based approximations to further training, for tabular, image, and text datasets and models. We find that the approximation quality of first-order methods is sometimes high but decays with the amount of further training. In contrast, the approximations given by influence function methods are more stable but surprisingly lower in quality.