Marginalized Importance Sampling for Off-Environment Policy Evaluation

Keyphrases

• Importance Sampling 100%
• Policy Evaluation 100%
• Environment Policy 100%
• Offline Data 100%
• Computationally Efficient 33%
• Occupancy 33%
• Estimation Error 33%
• Sampling Error 33%
• Two-step Procedure 33%
• Real-time Performance 33%
• Second Term 33%
• Real-world Deployment 33%
• Reinforcement Learning Algorithm 33%
• Robotic Arm 33%
• Error Propagation 33%
• First Term 33%
• Reinforcement Learning 33%
• Cart-pole 33%
• Reasonable Range 33%
• Policy Data 33%
• Robust Policy 33%
• Cheetah 33%
• Collection Policy 33%
• Direct Supervision 33%
• Indirect Supervision 33%
• Large Density Ratio 33%
• Sample Complexity 33%
• 7-DOF 33%
• Agent Policy 33%
• Gazebo Simulator 33%
• Learning Policy 33%
• Sim-to-real 33%

Computer Science

• Importance Sampling 100%
• Reinforcement Learning 66%
• Estimation Error 33%
• Robot 33%
• Collection Policy 33%