Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Evaluating Data Influence in Meta Learning
Chenyang Ren, Huanyi Xie, Shu Yang, Meng Ding, Dongrui Liu, Lijie Hu, Di Wang
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 task-IF (total gradient and Hessian) instead of direct IF (ablation) | MAML task removal · omniglot | accuracy: 0.5422 -> 0.7804 | Improved |
| use task-IF instead of direct IF (ablation) | MAML task removal · MNIST | accuracy: 0.3247 -> 0.6118 | Improved |
| use task-IF instead of direct IF (ablation) | MAML task removal · MINI-Imagenet | accuracy: 0.2214 -> 0.2668 | Improved |
| use direct IF instead of task-IF (faster, less accurate) (ablation) | MAML task removal · omniglot | runtime (s): 65.69 -> 7.46 | Improved |
| use task-IF instead of retraining | MAML task removal · omniglot | runtime (s): 316.74 -> 65.69 | Improved |
| use task-IF instead of retraining | MAML task removal · MINI-Imagenet | accuracy: 0.3071 -> 0.2668 | Hurt |
| use instance-IF (training) instead of retraining | MAML instance removal · omniglot | runtime (s): 251.76 -> 4.49 | Improved |
| use instance-IF (validation) instead of retraining | MAML instance removal · MNIST | accuracy: 0.6265 -> 0.5476 | Hurt |
| remove tasks by influence score instead of randomly | MAML with 80% flipped-label tasks · MNIST harmful task removal, 25% checked | test accuracy: 55 -> 62.5 | Improved |
| remove top-influence tasks instead of random tasks | MAML · MNIST task level | test accuracy: 0.62 -> 0.58 | Improved |
Closest papers by meaning
- Meta-Dataset: A Dataset of Datasets for Learning to Learn from Few Examples (ICLR 2020 · similarity 0.86)
- The Mirrored Influence Hypothesis: Efficient Data Influence Estimation by Harnessing Forward Passes (CVPR 2024 · similarity 0.86)
- Taming Hyperparameter Sensitivity in Data Attribution: Practical Selection Without Costly Retraining (NeurIPS 2025 · similarity 0.85)
- A Versatile Influence Function for Data Attribution with Non-Decomposable Loss (ICML 2025 · similarity 0.85)
- Capturing the Temporal Dependence of Training Data Influence (ICLR 2025 · similarity 0.85)
- Rescaled Influence Functions: Accurate Data Attribution in High Dimension (NeurIPS 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.
