Towards Understanding Asynchronous Advantage Actor-Critic: Convergence and Linear Speedup

Keyphrases

• Linear Speedup 100%
• Asynchronous Advantage Actor-critic 100%
• Convergence Guarantee 100%
• Convergence with Speed 100%
• Reinforcement Learning 66%
• Parallelization 66%
• Reinforcement Learning Algorithm 66%
• Sample Complexity 66%
• Asynchrony 33%
• Parallel Implementation 33%
• Global Convergence 33%
• Worker number 33%
• Numerical Test 33%
• Two-time Scale 33%
• Theoretical Properties 33%
• Markovian 33%
• Local Convergence 33%
• Synthetic Environment 33%
• Non-asymptotic Analysis 33%
• Asynchronous Implementation 33%
• Softmax 33%
• Atari Games 33%
• OpenAI Gym 33%
• A3C Algorithm 33%
• AC Algorithm 33%
• Popular 33%
• Performance Gains 33%
• Non-asymptotic Convergence 33%
• Policy Function Approximation 33%
• Gym Environment 33%
• Asynchronous Reinforcement Learning 33%

Mathematics

• Asymptotics 100%
• Local Convergence 100%
• Asymptotic Analysis 100%
• OpenAI 100%

Computer Science

• Reinforcement Learning 100%
• Linear Speedup 100%
• Parallelism 50%
• Approximation (Algorithm) 25%
• Performance Gain 25%
• Global Convergence 25%
• Parallel Implementation 25%
• Synthetic Environment 25%