Research map / Architectures and efficiency / Pruning and efficient tuning
Theoretical and Practical Perspectives on what Influence Functions Do
Andrea Schioppa, Katja Filippova, Ivan Titov, Polina Zablotskaia
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 |
|---|---|---|---|
| use Conjugate Residual instead of Conjugate Gradient for inverse Hessian-vector products in HIF (ablation) | Hessian influence function (HIF) · BERT/ResNet | Pearson correlation R(t) | Improved |
| retrain longer after the checkpoint (time since checkpoint increases) | influence scores (TracIn, HIF) · BERT SST2 and ResNet CIFAR10 | Pearson correlation R(t) | Hurt |
| Proponents-correction and Opponents-tuning instead of random relabeling baseline | BERT · SST2 | success rate of correcting mis-predictions | Improved |
| Proponents-correction instead of random relabeling baseline | ResNet · CIFAR10 | success rate: +2% | Improved |
| Proponents-correction instead of baseline | ResNet · CIFAR10 | steps to correct: 6% | Improved |
| Opponents-tuning instead of baseline | ResNet · CIFAR10 | success rate: 54% | No clear effect |
| fine-tune ResNet with SGD vs BERT with Adam; slower parameter divergence | influence scores · ResNet CIFAR10 | Pearson correlation R(t): 0.3 | Mixed |
Closest papers by meaning
- Towards Robust Influence Functions with Flat Validation Minima (ICML 2025 · similarity 0.91)
- Rescaled Influence Functions: Accurate Data Attribution in High Dimension (NeurIPS 2025 · similarity 0.87)
- Neural Networks for Learnable and Scalable Influence Estimation of Instruction Fine-Tuning Data (NeurIPS 2025 · 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 0.86)
- Capturing the Temporal Dependence of Training Data Influence (ICLR 2025 · similarity 0.86)
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.
