Independent Reading · 2024

Online Bayesian Persuasion Without a Clue

Francesco Bacchiocchi, Matteo Bollini, Matteo Castiglioni, Alberto Marchesi, and Nicola Gatti · NeurIPS

Bayesian PersuasionOnline LearningIncomplete Information

Full text

Read the Paper

Use View PDF or Download PDF to read the full paper on this device.

Paper Information

Authors
Francesco Bacchiocchi, Matteo Bollini, Matteo Castiglioni, Alberto Marchesi, and Nicola Gatti
Venue
NeurIPS · 2024
Open official paper

Summary

This paper studies repeated Bayesian persuasion when the sender knows neither the prior over states nor the receiver’s utilities.

Core Idea

A sender can learn effective signaling schemes from receiver responses even without an initial model of the surrounding environment.

Method

The authors develop an explore-then-commit algorithm over a non-standard representation of signaling schemes and prove matching regret and sample-complexity bounds.

Key Takeaways

  • Strategic information design can be learned under severe model uncertainty.
  • Receiver best-response feedback can reveal useful structure about unknown preferences.
  • The representation of signaling schemes is central to obtaining tight learning guarantees.

Questions

  • How robust is the method when receiver behavior is noisy rather than exactly optimal?
  • Can the approach adapt when the prior or receiver utilities change over time?