Incentivizing Exploration among Behavioral Agents with Unbiased Histories
summary
Model assumes two arms with binary rewards, agents estimate payoffs by empirical averages and see only sub histories chosen ex ante. Main claim is a deterministic signaling policy that forces exploration among behavioral frequentist agents and attains provable regret bounds. Construction uses two-level focus groups for initial exploration (analysis via concentration and anti-concentration) to get ~T^{2/3} regret, and a multi‑layer, information‑reusing refinement that approaches near‑optimal T^{1/2} regret, matching adaptive bandit rates.
AI-written summary · about
chapters (10)
- 0:00Intro
- 0:40incentivized exploration.
- 2:30multi-armed bandits.
- 3:57centralized planner.
- 13:00disparities.
- 17:18setting.
- 22:09behavioral model.
- 27:17design goal.
- 29:46first attempts.
- 36:533-level.
related talks
22mInterface Design Optimization as a Multi-Armed Bandit Problem
09mDecision Making under Miscalibration - 8 minute
21mDecision Making under Miscalibration
32mVirtual talk by Janardhan Kulkarni
09mAre Explanations Helpful? A Comparative Study of the Effects of Explanations in AI-Assisted ...
53mIterative preference learning methods for large language model post training
45mAgreement and Alignment for Human-AI Collaborative Decision Making
11mThe Unreasonable Effectiveness of Prompt Learning