Tight Certification of Adversarially Trained Neural Networks via Nonconvex Low-Rank Semidefinite Relaxations

Keyphrases

• Tight 100%
• Nonconvex 100%
• Low-rank 100%
• Training Model 100%
• Semidefinite Relaxation 100%
• Neural Network Training 100%
• Linear Programming Method 100%
• Non-convex Relaxation 100%
• Polynomial Time 50%
• Neural Network Model 50%
• Programming Method 50%
• Rank Constraint 50%
• Global Optimality 50%
• Closing the Gap 50%
• Non-convexity 50%
• Semidefinite Programming 50%
• Mixed Integer Linear Programming 50%
• Branch-and-bound Method 50%
• Convex Relaxation 50%
• Adversarial Perturbation 50%
• Quality Certification 50%
• Adversarial Training 50%
• Local Optimization Algorithm 50%

Computer Science

• Programming Method 100%
• Semidefinite Programming 100%
• Trained Neural Network 100%
• Linear Programming 100%
• Programming Technique 50%
• Neural Network Model 50%
• Optimization Algorithm 50%
• Polynomial Time 50%
• Global Optimality 50%
• Branch-and-Bound Algorithm Design 50%
• Adversarial Machine Learning 50%
• Mixed-Integer Linear Programming 50%
• Local Optimization 50%
• Convex Relaxation 50%
• Quality Certification 50%