A Theoretical Analysis of Fine-tuning with Linear Teachers

Keyphrases

• Sample Complexity 100%
• Linear Regression 66%
• Similarity Measure 33%
• Training Data 33%
• Deep Learning 33%
• Task Structure 33%
• Rectified Linear Unit (ReLU) 33%
• Realistic Settings 33%
• Complexity Reduction 33%
• Theoretical Understanding 33%
• Covariance Structure 33%
• Downstream Task 33%
• Inductive Bias 33%
• Gradient-based Training 33%

Computer Science

• Common Practice 100%
• Training Data 100%
• Complexity Reduction 100%
• Deep Learning Method 100%

Mathematics

• Linear Regression Analysis 100%
• Similarity Measure 50%
• Training Data 50%
• Covariance Structure 50%
• Deep Learning Method 50%