Research map / Label-efficient and robust learning / Noisy labels
Noise Accumulation in High Dimensional Classification and Total Signal Index
Miriam R. Elman, Jessica Minnier, Xiaohui Chang, Dongseok Choi
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 |
|---|---|---|---|
| increase sample size from n=200 to n=5000 (ablation) | Random forest · S1(1), q=5000 | median test error (%): 19.5 -> 8.3 | Improved |
| increase number of nonzero signal elements from m=5 to m=30 (ablation) | Random forest · S1(2), q=5000 | median test error (%): 31.3 -> 3.8 | Improved |
| scale signal and predictors together, m:p from 10:5000 to 90:45000 (ablation) | Random forest · S1(3) | median test error (%): 19.5 -> 1.5 | Improved |
| shrink nonzero element magnitude from 1 to sqrt(m) (ablation) | Random forest · S1(1) n=200, q=5000 | median test error (%): 19.5 -> 50.0 | Hurt |
| shrink nonzero element magnitude from 1 to sqrt(m) (ablation) | Random forest · S1(2) m=30, q=5000 | median test error (%): 3.8 -> 49.0 | Hurt |
| use SVM instead of RF or BCT (ablation) | Classifier · Scenario 3, q=5000 | median test error (%): 7.5 -> 39.5 | Hurt |
| add more noise predictors, q from <=10 to 5000, with RF | Random forest · Scenario 1 | median test error (%): 2.5 -> 0.0 | Improved |
| add more noise predictors, q from <=10 to 5000, with SVM | SVM · Scenario 2 | median test error (%): 1.5 -> 20.5 | Hurt |
| add more noise predictors, q from <=10 to 5000, with SVM | SVM · Scenario 4 | median test error (%): 24.5 -> 35.0 | Hurt |
| add more noise predictors with boosted trees | BCT · Scenario 4 | median test error (%): 25.5 -> 10.0 | Improved |
| reduce nonzero elements from m=10 to m=2 (weaker signal strength) | Random forest · q=5000 | median test error (%): 0.0 -> 7.5 | Hurt |
| reduce nonzero value from 3 to 1 with m=10 | Random forest · q=5000 | median test error (%): 0.0 -> 19.5 | Hurt |
Closest papers by meaning
- Identifying Hard Noise in Long-Tailed Sample Distribution (ECCV 2022 · similarity 0.86)
- Leveraging Data Symmetries to Select an Optimal Subset of Training Data under Label Noise (NeurIPS 2026 · similarity 0.86)
- Rethinking Benign Overfitting in Two-Layer Neural Networks (ICML 2025 · similarity 0.84)
- Noise Matters: Optimizing Matching Noise for Diffusion Classifiers (NeurIPS 2025 · similarity 0.84)
- Capturing the denoising effect of PCA via compression ratio (NeurIPS 2024 · similarity 0.84)
- BGDB: Bernoulli-Gaussian Decision Block with Improved Denoising Diffusion Probabilistic 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.
