Research map / Learning theory and algorithms / Matrix and tensor methods
Sparse Projection Oblique Randomer Forests
Tyler M. Tomita, James Browne, Cencheng Shen, Jaewon Chung, Jesse L. Patsolic, Benjamin Falk, Carey E. Priebe, Jason Yim, Randal Burns, Mauro Maggioni, Joshua T. Vogelstein
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 |
|---|---|---|---|
| replace axis-aligned splits with sparse random oblique projections | Random Forest · Consistency problem (p = 2) | error rate: lower bound of error 1/6 for Random Forest | Improved |
| replace axis-aligned splits with sparse random oblique projections | Random Forest · UCI hill_valley (5-fold CV) | Cohen's kappa (%): 12 -> 100 | Improved |
| replace axis-aligned splits with sparse random oblique projections | Random Forest · UCI hill_valley-noise (5-fold CV) | Cohen's kappa (%): 3 -> 90 | Improved |
| replace axis-aligned splits with sparse random oblique projections | Random Forest · UCI balance_scale (5-fold CV) | Cohen's kappa (%): 77 -> 94 | Improved |
| replace axis-aligned splits with sparse random oblique projections | Random Forest · UCI adult (5-fold CV) | Cohen's kappa (%): 82.42 -> 82.38 | No clear effect |
| use sparse random projections instead of Random Forest axis-aligned splits | Random Forest · Sparse parity (p = 20) | error rate | Improved |
| use SPORF instead of Forest-RC on the orthant problem | Forest-RC · Orthant (p = 6) | error rate | Improved |
| add many noise dimensions (10, 100, 1000) to UCI datasets, SPORF versus CCF and RerF | SPORF · UCI with Gaussian noise dimensions | mean Cohen's kappa | Improved |
| use default d = p, lambda = 3/p instead of tuned best d = p^2, lambda = 4/p | SPORF · UCI default hyperparameters | median rank of kappa: best d = p^2, lambda = 4/p; default d = p, lambda = 3/p | Mixed |
| sample d greater than p projections per node (ablation) | SPORF · Sparse parity (p = 20, n = 5000) | error rate and training time | Mixed |
| pack the trained forest (Forest Packing) for prediction | SPORF · MNIST, Higgs, p53 prediction time | prediction time: roughly ten times faster than XGBoost, over 100 times faster than standard SPORF | Improved |
| sample sparse lambda-density random matrices versus fixed-sparsity projections in Forest-RC (ablation) | SPORF · Sparse parity (n = 5000) and orthant (n = 400) | error rate vs lambda | Improved |
Closest papers by meaning
- Jacobian Aligned Random Forests (ICLR 2026 · similarity 0.82)
- Minimax Rates for High-Dimensional Random Tessellation Forests (JMLR 2024 · similarity 0.82)
- Debiasing Random Oblique Projections for Subsampled OLS and Fast CUR in High Dimensions (NeurIPS 2026 · similarity 0.81)
- Robust Visual Tracking Using Oblique Random Forests (CVPR 2017 · similarity 0.80)
- Geometry- and Accuracy-Preserving Random Forest Proximities (TPAMI 2023 · similarity 0.80)
- Randomization as Regularization: A Degrees of Freedom Explanation for Random Forest Success (JMLR 2020 · similarity 0.80)
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.
