Bayesian Influence Functions for Hessian-Free Data Attribution
This work addresses a bottleneck in data attribution for deep learning, offering a scalable solution for neural networks with billions of parameters.
The paper tackles the challenge of applying classical influence functions to deep neural networks by proposing a local Bayesian influence function that replaces Hessian inversion with loss landscape statistics estimated via stochastic-gradient MCMC sampling, achieving state-of-the-art results on predicting retraining experiments.
Classical influence functions face significant challenges when applied to deep neural networks, primarily due to non-invertible Hessians and high-dimensional parameter spaces. We propose the local Bayesian influence function (BIF), an extension of classical influence functions that replaces Hessian inversion with loss landscape statistics that can be estimated via stochastic-gradient MCMC sampling. This Hessian-free approach captures higher-order interactions among parameters and scales efficiently to neural networks with billions of parameters. We demonstrate state-of-the-art results on predicting retraining experiments.