regret analysis
A study of how well a decision-making algorithm performs compared to the best possible action it could have taken in hindsight. This analysis is important in reinforcement learning for evaluating and improving algorithm performance.
- A Novel General Framework for Sharp Lower Bounds in Succinct Stochastic Bandits
- Bandit and Delayed Feedback in Online Structured Prediction
- Contextual Dynamic Pricing with Heterogeneous Buyers
- Eluder dimension: localise it!
- Finite Sample Analyses for Continuous-time Linear Systems: System Identification and Online Control
- Improved Best-of-Both-Worlds Regret for Bandits with Delayed Feedback
- Improved Confidence Regions and Optimal Algorithms for Online and Offline Linear MNL Bandits
- Individual Regret in Cooperative Stochastic Multi-Armed Bandits
- MALinZero: Efficient Low-Dimensional Search for Mastering Complex Multi-Agent Planning
- Markov Persuasion Processes: Learning to Persuade From Scratch
- On the Universal Near Optimality of Hedge in Combinatorial Settings
- Online Two-Stage Submodular Maximization
- Pareto Optimal Risk-Agnostic Distributional Bandits with Heavy-Tail Rewards
- Precise Asymptotics and Refined Regret of Variance-Aware UCB
- Provably Efficient RL under Episode-Wise Safety in Constrained MDPs with Linear Function Approximation
- Variance-Aware Feel-Good Thompson Sampling for Contextual Bandits