Multicomposite nonconvex optimization for training deep neural networks

Mathematics

• Nonconvex Optimization 100%
• Training 89%
• Neural Networks 79%
• Semismooth Newton Method 58%
• Arbitrary 43%
• Stationary Solutions 42%
• Optimization Problem 30%
• Gradient Projection 29%
• Non-differentiability 29%
• Training Samples 28%
• Non-convexity 27%
• Activation Function 24%
• Enhancement 24%
• Majorization 24%
• Penalization 24%
• Framework 24%
• Stationarity 22%
• MATLAB 21%
• Neuron 21%
• Computational Cost 21%
• Knowledge 16%
• Costs 15%
• Numerical Results 13%
• Demonstrate 12%

Engineering & Materials Science

• Deep neural networks 81%
• Newton-Raphson method 61%
• Neurons 25%
• Chemical activation 25%
• MATLAB 22%
• Costs 10%