Research map / Learning theory and algorithms / Prediction and decision losses
Less Is Better: Unweighted Data Subsampling via Influence Function
Zifeng Wang, Hong Zhu, Zhenhua Dong, Xiuqiang He, Shao-Lun Huang
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 Sigmoid influence sampling instead of full-set training | Logistic regression · UCI Adult | average logloss: 0.3383 -> 0.3382 | Improved |
| Use Sigmoid influence sampling instead of full-set training | Logistic regression · cifar10 | average logloss: 0.6847 -> 0.6819 | Improved |
| Use Sigmoid influence sampling instead of full-set training | Logistic regression · Avazu-site | average logloss: 0.4499 -> 0.4486 | Improved |
| Use Linear influence sampling instead of full-set training | Logistic regression · Criteo1% | average logloss: 0.4763 -> 0.4755 | Improved |
| Use Linear influence sampling instead of full-set training | Logistic regression · Company | average logloss: 0.1955 -> 0.1952 | Improved |
| Use Sigmoid influence sampling instead of Dropout | Logistic regression · UCI breast-cancer | average logloss: 0.0785 -> 0.0803 | Hurt |
| Use Sigmoid influence sampling instead of Dropout | Logistic regression · diabetes | average logloss: 0.5083 -> 0.5068 | Improved |
| Use Sigmoid influence sampling instead of Dropout | Logistic regression · News20 | average logloss: 0.5072 -> 0.5075 | Hurt |
| Use Linear influence sampling instead of Dropout | Logistic regression · MNIST | average logloss: 0.0223 -> 0.0245 | Hurt |
| Use Linear influence sampling instead of Dropout | Logistic regression · Covertype | average logloss: 0.7745 -> 0.6872 | Improved |
| Use probabilistic influence sampling rather than deterministic Dropout (ablation) | UIDS · out-of-sample test across datasets | logloss | Mixed |
| Use Linear sampling rather than Sigmoid sampling (ablation) | UIDS · Criteo1% | average logloss: 0.4756 -> 0.4755 | Improved |
Closest papers by meaning
- Towards a statistical theory of data selection under weak supervision (ICLR 2024 · similarity 0.87)
- Most Influential Subset Selection: Challenges, Promises, and Beyond (NeurIPS 2024 · similarity 0.86)
- A Versatile Influence Function for Data Attribution with Non-Decomposable Loss (ICML 2025 · similarity 0.84)
- Boosting Test Performance with Importance Sampling--a Subpopulation Perspective (AAAI 2025 · similarity 0.83)
- Scaling laws for learning with real and surrogate data (NeurIPS 2024 · similarity 0.83)
- Group-robust Sample Reweighting for Subpopulation Shifts via Influence Functions (ICLR 2025 · similarity 0.83)
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.
