로그인
통찰 - Policy Optimization Algorithms
Reusing Historical Trajectories in Natural Policy Gradient via Importance Sampling: Convergence and Convergence Rate
The author explores the convergence and convergence rate of a natural policy gradient method with reusing historical trajectories via importance sampling, showing that it improves the convergence rate by an order of O(1/K).
Simple Policy Optimization: Improving Sample Efficiency and Stability in Reinforcement Learning
Introducing Simple Policy Optimization (SPO) as a more efficient and stable alternative to traditional Proximal Policy Optimization (PPO) algorithms in reinforcement learning.
