BEYOND LINEARIZATION: ON QUADRATIC AND HIGHER-ORDER APPROXIMATION OF WIDE NEURAL NETWORKS

Keyphrases

• Quadratic Orders 100%
• Linearization 100%
• High-order Approximation 100%
• Neural Tangent Kernel 100%
• Wide Neural Networks 100%
• Neural Network 60%
• Parametrized 40%
• Linearized Model 40%
• Higher-order Terms 20%
• Taylor Expansion 20%
• Empirical Performance 20%
• Neural Network Model 20%
• Simple Function 20%
• Taylor Series 20%
• Quadratic Model 20%
• Optimization Landscape 20%
• Random Networks 20%
• Distributional Hypothesis 20%
• Two-layer Network 20%
• Sample Complexity Bound 20%
• Randomization Method 20%

Mathematics

• Neural Network 100%
• Approximation Order 100%
• Taylor Expansion 20%
• Taylor Series 20%
• Simple Function 20%
• Quadratic Model 20%
• Distributional Assumption 20%
• Randomization 20%

Computer Science

• Neural Network 100%
• Approximation (Algorithm) 100%
• Network Layer 20%
• Quadratic Model 20%
• Taylor Series 20%
• Taylor Expansion 20%