Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Off-Policy Confidence Sequences
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use vector betting instead of scalar betting (ablation) | MOPE · synthetic environments; average confidence-sequence width over 500000 steps | confidence-sequence width | Helped |
| use separate bets for a grid of 200 v values (ablation) | MOPE · synthetic environments; average confidence-sequence width over 500000 steps | confidence-sequence width: 32 sec; 10440 sec | Mixed |
| maximize exact log wealth rather than its lower bound (ablation) | MOPE · synthetic environments; average confidence-sequence width over 500000 steps | confidence-sequence width: 32 sec; 15882 sec | No clear effect |
| include a bet on w-1 in addition to wr-v | MOPE confidence sequence · synthetic coverage experiment, alpha=0.05 | coverage | Helped |
| add a reward predictor | MOPE confidence sequence · MNIST8M, first 1 million samples | lower confidence bound | Helped |
| add a reward predictor | MOPE confidence sequence · MNIST8M, first 1 million samples | upper confidence bound | Hurt |
| combine predictor-based and predictor-free processes with double hedging | MOPE confidence sequence · MNIST8M, first 1 million samples | confidence-sequence width | Helped |
| use off-policy confidence sequences rather than an on-policy A/B test | gated deployment of the candidate policy · MNIST8M gated deployment | samples to decide candidate is better: 657 samples; 440 samples (220 for each policy) | Mixed |
Closest papers by meaning
- Empirical Likelihood for Contextual Bandits (NeurIPS 2020 · similarity 0.91)
- Optimal and Adaptive Off-policy Evaluation in Contextual Bandits (ICML 2017 · similarity 0.90)
- Conformal Off-Policy Prediction in Contextual Bandits (NeurIPS 2022 · similarity 0.90)
- Logarithmic Smoothing for Pessimistic Off-Policy Evaluation, Selection and Learning (NeurIPS 2024 · similarity 0.89)
- Bayesian Regret Minimization in Offline Bandits (ICML 2024 · similarity 0.87)
- Selective Uncertainty Propagation in Offline RL (AAAI 2025 · similarity 0.87)
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.