Research map / Reinforcement learning / Policy gradient and actor-critic
Learning to Bet for Horizon-Aware Anytime-Valid Testing
Ege Onur Taga, Samet Oymak, Shubhanshu Shekhar
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| learn betting policy with universal DQN instead of hedging over ε-greedy schedules or STaR-Bets | horizon-aware betting CS · Beta mixture, N=100, mu=0.25 | CS width at deadline: 0.079 -> 0.066 | Improved |
| learn betting policy with DQN instead of STaR-Bets | horizon-aware betting CS · Beta mixture, N=100, mu=0.4 | CS width at deadline: 0.141 -> 0.134 | Improved |
| learn betting policy with DQN instead of hedged ε-schedules | horizon-aware betting CS · Beta mixture, N=100, mu=0.65 | CS width at deadline: 0.085 -> 0.075 | Improved |
| expand DQN action space from 3 to 9 actions (ablation) | DQN betting policy · rejection probability by deadline | power | No clear effect |
| use DQN trained on Beta family on real data without fine-tuning | DQN betting policy · six real datasets (DNA methylation, humidity), N=100 | rejection probability by deadline: best in 5/6 cases | Improved |
| replace terminal-power reward with urgency reward (1 - t/N) (ablation) | DQN betting policy · Beta mixture, N=100 | power at t=N | Hurt |
| replace terminal-power reward with early-bonus reward 1+(1-t/N) (ablation) | DQN betting policy · Beta mixture, N=100 | power at t=N | Mixed |
| apply DQN trained on Beta to logit-normal data | DQN betting policy · logit-normal data | power | Improved |
| evaluate DQN at longer deadlines N in {250,300,350} | DQN betting policy · longer horizons | power | Improved |
| retrain DQN for alpha=0.01 | DQN betting policy · alpha=0.01 | power | Improved |
Closest papers by meaning
- Time-Sensitive Anytime-Valid Testing (NeurIPS 2026 · similarity 0.84)
- The Effective Horizon Explains Deep RL Performance in Stochastic Environments (ICLR 2024 · similarity 0.84)
- Test Where Decisions Matter: Importance-driven Testing for Deep Reinforcement Learning (NeurIPS 2024 · similarity 0.84)
- Long-Horizon Q-Learning: Accurate Value Learning via n-Step Inequalities (NeurIPS 2026 · similarity 0.83)
- Robustness Verification of Deep Reinforcement Learning Based Control Systems Using Reward Martingales (AAAI 2024 · similarity 0.83)
- Horizon-free Reinforcement Learning in Adversarial Linear Mixture MDPs (ICLR 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
