Robustness to Unbounded Smoothness of Generalized SignSGD

Keyphrases

• SignSGD 100%
• Clipping 60%
• Deep Learning 40%
• Noise Effects 20%
• Convergence Rate 20%
• Lipschitz 20%
• Non-adaptive 20%
• Mini 20%
• Recurrent Neural Network 20%
• Vanilla 20%
• Gradient Descent 20%
• Smoothness Assumption 20%
• Nonconvex Optimization 20%
• Relaxed Conditions 20%
• Learning Task 20%
• Smoothness Condition 20%
• Learning Objectives 20%
• Adam Algorithm 20%
• Gradient Norm 20%
• Adaptive Gradient Methods 20%

Mathematics

• Deep Learning 100%
• Objective Function 50%
• Convergence Rate 50%
• Neural Network 50%
• Learning Task 50%