Research map / Learning theory and algorithms / Prediction and decision losses
GEX: A flexible method for approximating influence via Geometric Ensemble
SungYub Kim, Kyungsu Kim, Eunho Yang
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 |
|---|---|---|---|
| remove linearization (use nonlinear Geometric Ensemble influence) (ablation) | IGEX · CIFAR-10 noisy label detection | AUC: 64.11 ± 0.34 -> 99.74 ± 0.02 | Improved |
| remove linearization (ablation) | IGEX · CIFAR-100 noisy label detection | AUC: 76.06 ± 0.36 -> 99.33 ± 0.03 | Improved |
| remove linearization (ablation) | IGEX · CIFAR-10-N relabeling | relabeled accuracy: 91.04 ± 0.16 -> 93.54 ± 0.05 | Improved |
| reduce ensemble size M_GE from 32 to 8 (ablation) | IGEX · CIFAR-10 noisy label detection | AP: 98.31 ± 0.06 -> 97.23 ± 0.11 | Hurt |
| reduce ensemble size M_GE from 32 to 4 (ablation) | IGEX · CIFAR-100 noisy label detection | AP: 96.08 ± 0.12 -> 92.99 ± 0.18 | Hurt |
| IGEX instead of CL | IGEX · CIFAR-10 noisy label detection | AUC: 57.60 ± 0.30 -> 99.74 ± 0.02 | Improved |
| IGEX instead of Deep-KNN | IGEX · CIFAR-100-N noisy label detection | AUC: 71.59 ± 0.21 -> 89.76 ± 0.01 | Improved |
| use IArnoldi instead of random projection IRandProj | IArnoldi · CIFAR-10 noisy label detection | AUC: 62.70 ± 0.19 -> 61.64 ± 0.13 | No clear effect |
| use checkpoint-based ITracInRP instead of IRandProj | ITracInRP · CIFAR-10 noisy label detection | AUC: 62.70 ± 0.19 -> 89.56 ± 0.14 | Improved |
| replace ITracInRP intermediate checkpoints with GE checkpoints (ablation) | ITracInRP · CIFAR-100 noisy label detection | AUC: 74.99 ± 0.25 -> 88.84 ± 0.22 | Improved |
| replace ITracInRP intermediate checkpoints with GE checkpoints (ablation) | ITracInRP · SVHN noisy label detection | AUC: 91.52 ± 0.16 -> 85.35 ± 0.20 | Hurt |
| relabel with Otsu threshold on CL self-influence in noisy real-world labels | CL · CIFAR-10-N relabeling | relabeled accuracy: 68.63 ± 0.32 -> 30.52 ± 0.02 | Hurt |
Closest papers by meaning
- Towards Robust Influence Functions with Flat Validation Minima (ICML 2025 · similarity 0.87)
- A Versatile Influence Function for Data Attribution with Non-Decomposable Loss (ICML 2025 · similarity 0.86)
- Influence Functions for Edge Edits in Non-Convex Graph Neural Networks (NeurIPS 2025 · similarity 0.86)
- Most Influential Subset Selection: Challenges, Promises, and Beyond (NeurIPS 2024 · similarity 0.85)
- DataInf: Efficiently Estimating Data Influence in LoRA-tuned LLMs and Diffusion Models (ICLR 2024 · similarity 0.85)
- Influence Functions for Scalable Data Attribution in Diffusion Models (ICLR 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.
