Maximum principle based algorithms for deep learning

Keyphrases

• Maximum Principle 100%
• Deep Learning 100%
• Principles-based 100%
• Gradient Method 50%
• Training Algorithm 50%
• Pontryagin Maximum Principle 50%
• Control Problem 25%
• Convergence Rate 25%
• Slow Manifold 25%
• Continuous-time 25%
• Optimality Conditions 25%
• Successive Approximation 25%
• Continuous Dynamical Systems 25%
• Error Estimates 25%
• Dynamical Systems Approach 25%
• Slow Convergence 25%
• Error Convergence 25%
• Trainable 25%
• Convergence Results 25%
• Flat Landscape 25%
• Alternative Training 25%

Mathematics

• Deep Learning Method 100%
• Maximum Principle 100%
• Gradient-Based Method 50%
• Control Problems 25%
• Saddle Point 25%
• Dynamical System 25%
• Slow Manifold 25%
• Successive Approximation 25%
• Continuous Time 25%
• Convergence Rate 25%
• Error Estimate 25%
• Convergence Result 25%
• Necessary Optimality Conditions 25%
• Initial Convergence 25%