Research map / Learning theory and algorithms / Prediction and decision losses
Stable Regression: On the Power of Optimization over Randomization
Dimitris Bertsimas, Ivan Paskov
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 |
|---|---|---|---|
| train on the hardest training/validation split found by optimization instead of a random split | L1 regression, unregularized · 12 datasets avg, 50/50 split | test MSE: 15.12% | Improved |
| same, with 90/10 split | L1 regression, unregularized · 12 datasets avg, 90/10 split | test MSE: 1.00% | Improved |
| train on hardest split instead of random split | Lasso · 12 datasets avg, 50/50 split | test MSE: 10.82% | Improved |
| compare against 5-fold and 10-fold cross validation | Lasso · 12 datasets avg | test MSE: 0.75% and 0.43% | Improved |
| optimized split instead of random split | unregularized regression · 12 datasets avg, 50/50 split | std of prediction error: 111.61% | Improved |
| optimized split instead of random split | Lasso · 12 datasets avg, 50/50 split | std of coefficients: 47.57% | Improved |
| optimized split instead of random split | Lasso · 12 datasets avg, 50/50 split | std of hyperparameter: 5.83% | Improved |
| optimized split instead of cross validation, known sparsity | L0-constrained regression · synthetic support recovery | accuracy and false alarm rate | Improved |
| optimized split instead of cross validation, estimated sparsity | L0-constrained regression · synthetic support recovery | accuracy | Mixed |
| use the optimization procedure instead of random splitting | training procedure · running time, 50/50 split | time: 2.0601 times longer | Hurt |
Closest papers by meaning
- Consistent Robust Regression (NeurIPS 2017 · similarity 0.90)
- Set Valued Predictions For Robust Domain Generalization (ICML 2025 · similarity 0.87)
- Understanding Domain Generalization: A Noise Robustness Perspective (ICLR 2024 · similarity 0.86)
- Structure-informed Risk Minimization for Robust Ensemble Learning (ICML 2025 · similarity 0.86)
- Piecewise Constant and Linear Regression Trees: An Optimal Dynamic Programming Approach (ICML 2024 · similarity 0.86)
- Distributionally Robust Feature Selection (NeurIPS 2025 · similarity 0.86)
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.
