Research map / Learning theory and algorithms / Prediction and decision losses
If Influence Functions are the Answer, Then What is the Question?
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| compare influence estimates to PBRF instead of cold-start retraining | Influence functions (MLP) · MNIST test loss difference | Pearson correlation: -0.55 -> 0.98 | Helped |
| compare influence estimates to PBRF instead of cold-start retraining | Influence functions (ResNet-20) · MNIST test loss difference | Pearson correlation: 0.09 -> 0.81 | Helped |
| compare influence estimates to PBRF instead of warm-start retraining | Influence functions (VGG13) · MNIST test loss difference | Pearson correlation: -0.28 -> 0.98 | Helped |
| increase network width (ablation) | Influence function discrepancy · MLP on 10% MNIST | linearization error | Helped |
| increase network depth (ablation) | Influence function discrepancy · MLP on 10% MNIST | gap components | No clear effect |
| increase training epochs (ablation) | Influence function discrepancy · MLP on 10% MNIST | non-convergence gap | Helped |
| increase weight decay (ablation) | Influence function discrepancy · MLP on 10% MNIST | overall discrepancy | Helped |
| increase damping (ablation) | Influence function discrepancy · MLP on 10% MNIST | solver/linearization error vs proximity gap | Mixed |
| remove more training examples (ablation) | Influence function discrepancy · MLP on 10% MNIST | linearization error and warm-start gap | Hurt |
Closest papers by meaning
- Influence Functions for Edge Edits in Non-Convex Graph Neural Networks (NeurIPS 2025 · similarity 0.89)
- Towards Robust Influence Functions with Flat Validation Minima (ICML 2025 · similarity 0.89)
- A Versatile Influence Function for Data Attribution with Non-Decomposable Loss (ICML 2025 · similarity 0.88)
- Influence Functions for Scalable Data Attribution in Diffusion Models (ICLR 2025 · similarity 0.87)
- Rescaled Influence Functions: Accurate Data Attribution in High Dimension (NeurIPS 2025 · similarity 0.86)
- Neural Networks for Learnable and Scalable Influence Estimation of Instruction Fine-Tuning Data (NeurIPS 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.