Research map / Learning theory and algorithms / Kernels and regression theory
Learning Optimal Linear Regularizers
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Tune regularization hyperparameters with the LP-based TuneReg procedure instead of random search or GP-EI-MCMC | regularization hyperparameter tuning · Coins, MNIST softmax regression, and Inception-v3 transfer learning | training runs to reach comparable accuracy: at least an order of magnitude more training runs | Helped |
| Choose L2 regularization rather than L1 for MNIST softmax regression | MNIST softmax regression · MNIST softmax regression | min. test loss: 0.0249 -> 0.0233 | Helped |
| Choose L2 regularization rather than label smoothing | MNIST softmax regression · MNIST softmax regression | min. test loss: 0.0254 -> 0.0233 | Helped |
| Use dropout rather than L2 regularization | MNIST softmax regression · MNIST softmax regression | max. test accuracy: 99.31% -> 99.34% | Helped |
| Choose L2 regularization rather than L1 | Inception-v3 transfer learning · Inception-v3 transfer learning | min. test loss: 0.309 -> 0.285 | Helped |
| Choose L2 regularization rather than label smoothing | Inception-v3 transfer learning · Inception-v3 transfer learning | min. test loss: 0.366 -> 0.285 | Helped |
| Use L2 regularization rather than dropout | Inception-v3 transfer learning · Inception-v3 transfer learning | max. test accuracy: 90.46% -> 90.74% | Helped |
| Use dropout rather than L2 regularization | MNIST softmax regression · MNIST softmax regression | min. test loss: 0.0233 -> 0.0463 | Hurt |
| Use dropout rather than L2 regularization | Inception-v3 transfer learning · Inception-v3 transfer learning | min. test loss: 0.285 -> 0.506 | Hurt |
| Use the LogitBeta regularizer for coin-bias estimation | coin-bias estimation · Coins | max. slack: 0 | Helped |
Closest papers by meaning
- Optimizing Nondecomposable Data Dependent Regularizers via Lagrangian Reparameterization Offers Significant Performance and Efficiency Gains (AAAI 2020 · similarity 0.88)
- Scalable Gradient-Based Tuning of Continuous Regularization Hyperparameters (ICML 2016 · similarity 0.88)
- Regularization Matters: Generalization and Optimization of Neural Nets v.s. their Induced Kernel (NeurIPS 2019 · similarity 0.87)
- Local Regularizer Improves Generalization (AAAI 2020 · similarity 0.87)
- A Statistical Theory of Regularization-Based Continual Learning (ICML 2024 · similarity 0.87)
- New Bounds for Hyperparameter Tuning of Regression Problems Across Instances (NeurIPS 2023 · similarity 0.87)
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.