Research map / Bandits and online learning / Contextual bandits and Thompson sampling
On Multi-Armed Bandit Designs for Dose-Finding Trials
Maryam Aziz, Emilie Kaufmann, Marie-Karelle Riviere
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 |
|---|---|---|---|
| add Thompson Sampling with logistic model in place of CRM | TS · MTD identification, 9 scenarios | recommendation of the MTD: 3 out of 9 | Mixed |
| add epsilon regularization to Thompson Sampling (ablation) | TS(epsilon) · MTD identification, 9 scenarios | recommendation of the MTD vs CRM: 5 out of 9 | Mixed |
| add admissible set to Thompson Sampling (ablation) | TS A · MTD identification, 9 scenarios | recommendation of the MTD vs CRM: 5 out of 9 | Mixed |
| exploit increasing-toxicity structure via logistic model instead of independent uniform priors | TS · MTD identification, n=36 | recommendation of the MTD: 37.6 -> 78.9 (Scenario 1) | Improved |
| add admissible set A or epsilon regularity to Thompson Sampling (ablation) | TS A and TS(epsilon) · MTD identification | allocation to high-toxicity doses: consistently reduces the allocation percentage of higher toxicity doses | Improved |
| add admissible set to Thompson Sampling (ablation) | TS A · MTD identification Sc. 3 | allocation % to MTD (dose 6): 43.2 -> 32.1 | Hurt |
| use 3+3 design instead of model-based designs | 3+3 · MTD identification | dose recommendation: 35.2 vs 77.2 (Scenario 1, dose 1) | Hurt |
| vary epsilon in TS(epsilon) over 0.02, 0.05, 0.1 (ablation) | TS(epsilon) · MTD identification | recommendation performance: 0.02, 0.05, 0.1 comparable | No clear effect |
| vary c1 in TS A over 0.6, 0.8, 0.9 (ablation) | TS A · MTD identification | recommendation performance: 0.6, 0.8, 0.9 comparable | No clear effect |
| replace MTA-RA with TS A | TS A · MED identification, scenarios 1-4 and 12-13 | optimal decision rate vs MTA-RA: 5 out of 6 | Improved |
| replace MTA-RA with TS | TS · MED identification, scenarios 1-4 and 12-13 | optimal decision rate vs MTA-RA: 4 out of 6 | Improved |
| add admissible set to Thompson Sampling (ablation) | TS A · MED identification | allocation to doses with toxicity above theta: always smaller for TS A than for MTA-RA | Improved |
Closest papers by meaning
- Safe Exploration in Dose Finding Clinical Trials with Heterogeneous Participants (ICML 2024 · similarity 0.90)
- A Bandit Approach to Sequential Experimental Design with False Discovery Control (NeurIPS 2018 · similarity 0.88)
- DARTS: Targeting Prognostic Covariates in Budget-Constrained Sequential Experiments (NeurIPS 2026 · similarity 0.85)
- Probabilistic Factorial Experimental Design for Combinatorial Interventions (ICML 2025 · similarity 0.84)
- The Choice of Noninformative Priors for Thompson Sampling in Multiparameter Bandit Models (AAAI 2024 · similarity 0.84)
- Data-driven Design of Randomized Control Trials with Guaranteed Treatment Effects (ICML 2025 · 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.
