Distributional Training Data Attribution: What do Influence Functions Sample?
This addresses the issue of accounting for training randomness in data attribution for deep learning practitioners, offering a novel perspective on IFs with practical applications.
The paper tackles the problem of randomness in deep learning training by introducing distributional training data attribution (d-TDA) to predict how model output distributions depend on the dataset, showing that influence functions (IFs) are inherently distributional and demonstrating utility in experiments like improving data pruning for vision transformers and identifying influential examples with diffusion models.
Randomness is an unavoidable part of training deep learning models, yet something that traditional training data attribution algorithms fail to rigorously account for. They ignore the fact that, due to stochasticity in the initialisation and batching, training on the same dataset can yield different models. In this paper, we address this shortcoming through introducing distributional training data attribution (d-TDA), the goal of which is to predict how the distribution of model outputs (over training runs) depends upon the dataset. Intriguingly, we find that influence functions (IFs), a popular data attribution tool, are 'secretly distributional': they emerge from our framework as the limit to unrolled differentiation, without requiring restrictive convexity assumptions. This provides a new perspective on the effectiveness of IFs in deep learning. We demonstrate the practical utility of d-TDA in experiments, including improving data pruning for vision transformers and identifying influential examples with diffusion models.