Idea

More extension on Multi-player Online Learning

• Multi-player bandit XX
  Multi-player linear bandit/RL/ContextualBandit?

• Fundamental problems
  Distributed algorithm
  Restricted communication
  Mutual influence

Multi-player MAB in Other Applications

• Multi-player MAB in sharing economy
  Movement cost
  Random supply of customers

• Multi-player MAB in Edge computing
  Mobility of users

• Multi-player MAB in crowdsourcing
  Workers on an arm needs to collaborate
  Skill complement?

• Multi-player MAB in Multi-path TCP

• Multi-player MAB in distributed cloud system like Alibaba

• Multi-player MAB in Recommender Systems

Sharing Economy Inspired Multi-player MAB

• Cooperative Multi-player MAB
  Divide reward using the optimal reward as a value function
  Use shapley value to divide the reward
  Can be used to model the centralized multi-player MAB problem
  

• 2021.03.16: the assignment problem (PPT)
  Two important factors: (1) restriction of accessing arms; (2) switch or movement cost
  

• 2021.03.15: Discussion with John (PPT)
  Comment1: switch or movement cost
  Comment2: each player is restricted a subset of arms
  Comment3: different penalty factors for wasting players and wasting demands respectively
  Comment4: change of location after delivering
  Comment5: preference among players and arms

• Multi-player MAB
  Key element 1: distributed learning of the model
  What information can assist distributed learning?
  What's the fundamental challenges in distributed learning?
  
  Key element 2: distributed coordination to the optimal solution
  What's the fundamental challenge in distributed coordination?
  What information can assist distributed coordination?
  
  Key element 3: incentive to induce collaboration or competition
  The challenge is that this significantly affect the information or singal in distributed learning or distributed coordination.