A Branch and Bound Framework for Stronger Adversarial Attacks of ReLU Networks

Keyphrases

• Activation Spaces 60%
• Adversarial Attack 100%
• Adversarial Examples 80%
• Attack Framework 20%
• Beam Search 20%
• Bounds Propagation 20%
• Branch-and-bound 100%
• Deep Neural Network 20%
• Diving 20%
• GPU-based 20%
• Gradient Descent 20%
• Hard Examples 20%
• Input Space 60%
• Integer Programming 20%
• Large Neighborhood Search 20%
• Mixed Integer Programming 40%
• Networking of Theories 20%
• Neural Network Verification 20%
• Non-convexity 20%
• Problem Instances 20%
• Programming Solver 40%
• Programming Theory 20%
• Propagation Method 20%
• ReLU Networks 100%
• Search Approach 20%
• Small Networks 20%
• Success Rate 20%

Computer Science

• Adversarial Example 80%
• Adversarial Machine Learning 100%
• Branch-and-Bound Algorithm Design 100%
• Deep Neural Network 20%
• Gradient Descent 20%
• Integer Programming 20%
• Mixed Integer Programming 40%
• Neural Network 20%
• Problem Instance 20%
• Search Neighborhood 20%