A Dynamic Observation Strategy for Multi-agent Multi-armed Bandit Problem

Udari Madhushani, Naomi Ehrich Leonard

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

We define and analyze a multi-agent multi-armed bandit problem in which decision-making agents can observe the choices and rewards of their neighbors under a linear observation cost. Neighbors are defined by a network graph that encodes the inherent observation constraints of the system. We define a cost associated with observations such that at every instance an agent makes an observation it receives a constant observation regret. We design a sampling algorithm and an observation protocol for each agent to maximize its own expected cumulative reward through minimizing expected cumulative sampling regret and expected cumulative observation regret. For our proposed protocol, we prove that total cumulative regret is logarithmically bounded. We verify the accuracy of analytical bounds using numerical simulations.

Original languageEnglish (US)
Title of host publicationEuropean Control Conference 2020, ECC 2020
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages1677-1682
Number of pages6
ISBN (Electronic)9783907144015
StatePublished - May 2020
Event18th European Control Conference, ECC 2020 - Saint Petersburg, Russian Federation
Duration: May 12 2020May 15 2020

Publication series

NameEuropean Control Conference 2020, ECC 2020

Conference

Conference18th European Control Conference, ECC 2020
CountryRussian Federation
CitySaint Petersburg
Period5/12/205/15/20

All Science Journal Classification (ASJC) codes

  • Artificial Intelligence
  • Decision Sciences (miscellaneous)
  • Control and Systems Engineering
  • Mechanical Engineering
  • Computational Mathematics
  • Control and Optimization

Fingerprint Dive into the research topics of 'A Dynamic Observation Strategy for Multi-agent Multi-armed Bandit Problem'. Together they form a unique fingerprint.

Cite this