Lingjie Duan, Takeshi Kubo, Kohei Sugiyama, Jianwei Huang, Teruyuki Hasegawa, Jean Walrand Singapore SUTD, KDDI, CHUK, UC Berkeley
Collaboration Models Lingjie Duan, Takeshi Kubo, Kohei Sugiyama, - - PowerPoint PPT Presentation
Collaboration Models Lingjie Duan, Takeshi Kubo, Kohei Sugiyama, - - PowerPoint PPT Presentation
Collaboration Models Lingjie Duan, Takeshi Kubo, Kohei Sugiyama, Jianwei Huang, Teruyuki Hasegawa, Jean Walrand Singapore SUTD, KDDI, CHUK, UC Berkeley DIMACS Workshop on the Economic Aspects of information Sharing 2/7/13 Overview Data
SLIDE 1
SLIDE 2
Overview
2
Data collection & cooperation on tasks Incentives for collaboration? We model cooperation in two situations: Model 1:
“Data Collection Game” : share reward if successful
Model 2:
“Task collaboration”: contract
SLIDE 3
Model 1
3
Cj Client Value V If n ≥ n0 collaborations Collaboration costs: iid, F(.) known C1 CN N known
SLIDE 4
Model 1
4
Cj Client Value V If n ≥ n0 collaborations Collaboration costs C1 CN Scheme: Client proposes a reward R R is divided among collaborators, if n ≥ n0 Questions: Whether to collaborate? How to choose R?
SLIDE 5
Model 1
5
Cj Client Value V If n ≥ n0 collaborations Collaboration costs C1 CN Scheme: Client proposes a reward R R is divided among collaborators Nash equilibrium: Collaborate if Ci ≤ γ γ = unique solution of
= 0 where m = B(N, F(γ))
SLIDE 6
Model 1
6
Scheme: Client proposes a reward R R is divided among collaborators N = 100 n0 = 40 Ci = U[0, 4] E.g., R = 100 →γ = 2 note that E[n] = 50
SLIDE 7
Model 1
7
Scheme: Client proposes a reward R R is divided among collaborators Choosing R:
where n = B(N, F(γ∗(R)))
SLIDE 8
Model 1
8
Scheme: Client proposes a reward R R is divided among collaborators Choosing R: V = 100 n0 = 30 Ci = U[0, 3]
SLIDE 9
Model 2
9
Kj Client Collaboration costs, per unit of effort K1 KN Reward r Effort t Client’s utility: Type i:
SLIDE 10
Model 2
10
Kj Client K1 KN Proposes
contract
Selects j : I will produce tm units of effort for reward rm
SLIDE 11
Model 2
11
Client N users, each user has type i w.p. qi Algorithm for optimal design of contract
SLIDE 12
Model 2
12
N = 120 3 types qi = 1/3 θi = 5
SLIDE 13
Summary
13
Design of collaborations 1) “Data Collection” Share R if n ≥ n0 V for client Collaborate if cost ≤ γ* Calculate optimal R 2) “Collaboration on task” User of type i w.p. qi Design of optimal contract
SLIDE 14
14 This presentation is based on the following paper: Lingjie Duan, Takeshi Kubo, Kohei Sugiyama, Jianwei Huang, Teruyuki Hasegawa, Jean Walrand, “Incentive Mechanisms for Smartphone Collaboration in Data Acquisition and Distributed Computing,” INFOCOM 2012.