Training Certifiably Robust Neural Networks with Efficient Local Lipschitz Bounds

Keyphrases

• Lipschitz Bounds 100%
• Robust Neural Networks 100%
• Local Lipschitz 100%
• Neural Network 75%
• Activation Function 75%
• Tight 50%
• Weight Matrix 50%
• Lipschitz Constant 50%
• Training Algorithm 50%
• Rectified Linear Unit (ReLU) 50%
• Certified Accuracy 50%
• Popular 25%
• State-of-the-art Techniques 25%
• Sparsity 25%
• Network Architecture 25%
• Tight Bounds 25%
• Induced Norms 25%
• Non-convexity 25%
• Safety-critical Applications 25%
• Max-min 25%
• Deep Neural Network 25%
• Trainable 25%
• Global Bounds 25%
• Plug-in Module 25%
• MNIST 25%
• Upper Threshold 25%
• CIFAR-10 25%
• Certified Robustness 25%

Computer Science

• Neural Network 100%
• Activation Function 75%
• training algorithm 50%
• Sparsity 25%
• Network Architecture 25%
• Critical Application 25%
• Desirable Property 25%
• Deep Neural Network 25%