Off-policy evaluation and learning from logged bandit feedback: Error reduction via surrogate policy

Keyphrases

• Error Reduction 100%
• Policy Learning 100%
• Feedback Error 100%
• Bandit Feedback 100%
• Inverse Propensity Score 100%
• Off-policy Evaluation 100%
• Maximum Likelihood 80%
• Historical Policy 60%
• Mean Squared Error 40%
• Batch Learning 40%
• Online Learning Environment 20%
• Computational Challenges 20%
• Non-asymptotic 20%
• Unseen 20%
• Training Data 20%
• Learning Settings 20%
• Exponential Model 20%
• Platform System 20%
• Policy-based 20%
• Online Systems 20%
• Interactive Systems 20%
• Supervised Learning 20%
• Largest Mean 20%
• Statistical Challenges 20%
• Recommendation System 20%
• Ad Placement 20%
• Multi-label Classification Problem 20%
• Asymptotically Unbiased 20%
• Good Action 20%
• Policy Training 20%
• Ad Recommendation 20%
• Action Context 20%
• Bandit Learning 20%
• Error Reduction Technique 20%
• Ad Platforms 20%

Mathematics

• Maximum Likelihood 100%
• Squared Error 50%
• Batch Learning 50%
• Classification Problem 25%
• Training Data 25%
• Unbiasedness 25%

Computer Science

• maximum-likelihood 100%
• Training Data 25%
• Classification Problem 25%
• Online Learning 25%
• Supervised Learning 25%
• Interactive System 25%