Research map / Label-efficient and robust learning / Recognition and long-tail
A flexible model-free prediction-based framework for feature ranking
Jingyi Jessica Li, Yiling Elaine Chen, Xin Tong
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 s-NPC with small alpha .01 instead of s-CC | feature ranking · Toy example | frequency feature 2 ranked top: 22.0% -> 98.4% | Improved |
| raise NP level alpha from .01 to .20 | s-NPC · Toy example | frequency feature 1 ranked top: 1.6% -> 99.0% | Mixed |
| use s-NPC or s-CC instead of Pearson correlation | marginal ranking · S6 two-feature model | frequency recovering true rank: 0% -> 100% | Improved |
| use s-CC instead of two-sample Wilcoxon | marginal ranking · S6 two-feature model | frequency recovering true rank: 0% -> 100% | Improved |
| use s-NPC (alpha=.10) instead of distance correlation | marginal ranking · S6 two-feature model | frequency recovering true rank: 0.5% -> 99.3% | Improved |
| increase sample size from 400 to 1000 | s-CC and s-NPC · S1 Gaussian | agreement with true ranks | Improved |
| decrease alpha (e.g. to 0.05) (ablation) | s-NPC · S1 Gaussian N=400 | rank variability | Hurt |
| use s-NPC instead of s-CC or SVM variants under sampling bias (class 1 proportion .5 -> .1) | feature ranking · S3 Gaussian, sampling bias | separating 10 informative features | Improved |
| use s-NPC instead of its SVM variant | s-NPC · S5 d=10000, N=200 | ranking accuracy of top 10 features | Improved |
| use alpha=.30 instead of .10 or .20 | s-NPC · S5 d=10000, N=200 | average rank of 10 informative features | Improved |
| use larger alpha .2 or .3 instead of .05 or .1 | s-NPC · S4 d=500, N=400 | ranking features 8-10 | Improved |
| use mean decrease in accuracy instead of mean decrease in Gini | random forest importance · S5 d=10000, N=200 | rank of 10th feature | Improved |
Closest papers by meaning
- TreeGrad-Ranker: Feature Ranking via $O(L)$-Time Gradients for Decision Trees (ICLR 2026 · similarity 0.85)
- Nonparametric Regression with Comparisons: Escaping the Curse of Dimensionality with Ordinal Information (ICML 2018 · similarity 0.85)
- SAND: One-Shot Feature Selection with Additive Noise Distortion (ICML 2025 · similarity 0.84)
- Estimating Conditional Mutual Information for Dynamic Feature Selection (ICLR 2024 · similarity 0.83)
- A Differentiable Rank-Based Objective for Better Feature Learning (ICLR 2025 · similarity 0.83)
- Accurate Estimation of Feature Importance Faithfulness for Tree Models (AAAI 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.
