Batches Stabilize the Minimum Norm Risk in High-Dimensional Overparametrized Linear Regression

Keyphrases

• Linear Regression 100%
• Minimum Norm 100%
• Quadratic Risk 40%
• Overparametrization 40%
• Learning Algorithm 20%
• Risk Behavior 20%
• Interpolation Points 20%
• Linear Regression Model 20%
• Noise Level 20%
• Combined Technique 20%
• Computational Efficiency 20%
• Computational Performance 20%
• Machine Learning Applications 20%
• Noisy Projections 20%
• Novel Combinations 20%
• Normal Approximation 20%
• Small Batch 20%
• Optimal Batch Size 20%
• Wasserstein Distance 20%
• Feature Overlap 20%
• Implicit Regularization 20%
• Double Descent 20%
• Random Subspace 20%
• Gaussian Feature 20%

Mathematics

• Linear Regression 100%
• Quadratic Risk 100%
• Gaussian Distribution 50%
• Regularization 50%
• Linear Regression Model 50%
• Normal Approximation 50%
• Batch Size 50%
• Interpolation Point 50%