Maximum principle based algorithms for deep learning

Mathematics

• Training Algorithm 82%
• Maximum Principle 72%
• Learning 69%
• Gradient 38%
• Slow Manifold 34%
• Trapping 32%
• Alternatives 31%
• Discrete Variables 30%
• Successive Approximation 28%
• Necessary Optimality Conditions 28%
• Training 26%
• Saddlepoint 26%
• Attack 25%
• Convergence Results 22%
• Convergence Rate 21%
• Error Estimates 20%
• Dynamical system 18%
• Iteration 16%
• Demonstrate 15%
• Framework 14%

Engineering & Materials Science

• Maximum principle 100%
• Deep learning 68%
• Hamiltonians 30%
• Dynamical systems 20%