Research map / Optimization / Optimizers
Bayesian Influence Functions for Hessian-Free Data Attribution
Philipp Alexander Kreer, Wilson Wu, Maxwell Adam, Zach Furman, Jesse Hoogland
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace Hessian inversion with local posterior covariance (local BIF) instead of EK-FAC | ResNet-9 on CIFAR-10 · CIFAR-10 LDS, small retrain subset | LDS | Improved |
| use local BIF instead of EK-FAC in a language model finetuning setting | Pythia-14M · Pythia-14M finetuning LDS | LDS | Hurt |
| use local BIF instead of EK-FAC for small image models | ResNet-9 / CIFAR-10 · compute time | runtime: EK-FAC around five times faster | Hurt |
| use local BIF instead of EK-FAC for large models | Pythia suite · Pythia scaling, evaluation time | evaluation time: 2 orders of magnitude faster at largest sizes | Improved |
| use local BIF instead of EK-FAC for GPU memory | Pythia suite · Pythia scaling, peak GPU RAM | peak GPU RAM | No clear effect |
| decrease attribution set size (ablation) | all TDA methods on CIFAR-10 · CIFAR-10 LDS | LDS | Improved |
| use RMSprop-SGLD preconditioning (ablation) | local BIF sampler · hyperparameter robustness | result quality | Improved |
Closest papers by meaning
- Influence Functions for Scalable Data Attribution in Diffusion Models (ICLR 2025 · similarity 0.87)
- A Study of Bayesian Neural Network Surrogates for Bayesian Optimization (ICLR 2024 · similarity 0.86)
- LoRIF: Low-Rank Influence Functions for Scalable Training Data Attribution (NeurIPS 2026 · similarity 0.86)
- A Versatile Influence Function for Data Attribution with Non-Decomposable Loss (ICML 2025 · similarity 0.86)
- Outlier Gradient Analysis: Efficiently Identifying Detrimental Training Samples for Deep Learning Models (ICML 2025 · similarity 0.85)
- Towards Robust Influence Functions with Flat Validation Minima (ICML 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
