Improved Linear Convergence of Training CNNs with Generalizability Guarantees: A One-Hidden-Layer Case

Keyphrases

• Neural Network 100%
• Hidden Layer 100%
• Gradient Method 100%
• Learning Problems 100%
• Linear Convergence 100%
• Convolutional Neural Network 66%
• Non-convexity 66%
• Rectified Linear Unit (ReLU) 66%
• Accelerated Gradient Method 66%
• Sample Complexity 66%
• Convergence Rate 33%
• Nonsmooth 33%
• Noise Level 33%
• Non-overlapping 33%
• Training Set 33%
• Training Data 33%
• Estimation Error 33%
• Initialization Method 33%
• Additive Noise 33%
• Gaussian Distribution 33%
• Model Estimation 33%
• Finite number 33%
• Output Noise 33%
• Tight Bounds 33%
• Global Minimizer 33%
• Linear Growth Rate 33%
• Nonconvex Learning 33%
• Number of Training Samples 33%
• Vanilla Gradient Descent 33%
• Squared Error Loss Function 33%

Computer Science

• Neural Network 100%
• Convolutional Neural Network 100%
• Learning Problem 100%
• Generalizability 100%
• Gradient Descent 100%
• Activation Function 66%
• ReLU Function 66%
• Convergence Rate 33%
• Training Data 33%
• Training Sample 33%
• Estimation Error 33%
• Infinite Number 33%
• Initialization Method 33%

Mathematics

• Neural Network 100%
• Generalizability 100%
• Linear Convergence 100%
• Convolutional Neural Network 100%
• Convergence Rate 33%
• Gaussian Distribution 33%
• Tensor 33%
• Training Data 33%
• Additive Noise 33%
• Infinite Number 33%
• Loss Function 33%
• Training Set 33%
• Finite Number 33%
• Training Sample 33%
• Gradient-Based Method 33%