Distributed Adversarial Training to Robustify Deep Neural Networks at Scale

Mathematics

• Training 100%
• Neural Networks 88%
• Attack 38%
• Distributed Optimization 29%
• Scale-up 14%
• Optimization Theory 11%
• Batch 10%
• Min-max 10%
• Stationary point 10%
• Speedup 9%
• Compression 9%
• Learning 8%
• Rate of Convergence 7%
• Gradient 7%
• Perturbation 6%
• First-order 6%
• Context 5%
• Demonstrate 5%
• Framework 5%
• Standards 5%
• Model 3%

Engineering & Materials Science

• Deep neural networks 91%
• Labeled data 36%