CHAINME

Transcript

1. Social Value Propagation for Supply Chain Formation by Toni Penya-Alba, Meritxell Vinyals, Jesus Cerquides and Juan A. Rodriguez-Aguilar for the 6th International OPTMAS Workshop 5/23/12 logoiiia-‐‑gran.png (1200×467) 1/1 www.iiia.csic.es/tic/sites/default/files/templates-‐‑logos/logoiiia-‐‑gran.png
2. The problem
3. Alice Carol Bob Dave Eve $8 $7 -$5 -$3 -$2
4. The Supply Chain Formation problem is that of finding the feasible configuration with maximum value. Alice Carol Bob Dave Eve $8 $7 -$5 -$3 -$2
5. Supply Chain Value = $8 - $5 - $2 = $1 Alice Carol Eve $8 -$5 -$2
6. Decentralized algorithm for SCF based on social value propagation. CONTRIBUTION BENEFITS Reduces computation. Reduces communication. Produces (near)optimal solutions.
7. Related work
8. SUPPLY CHAIN FORMATION TIMELINE Decentralized High communication Low memory & computation High valued solutions Centralized Optimal Multi-unit Low scalability Decentralized Low communication Low memory & computation Low valued solutions SAMP-SB MMUCA RB-LBP 2003 2008 2012
9. Social Value Propagation
10. PARTICIPANT’S SOCIAL VALUE Frank $5 Eve $6 Bob -$3 Alice -$2 Dave -$5 Carol -$4 Hank $1 Gene $2
11. PARTICIPANT’S SOCIAL VALUE Social value for participant p how much better is for the other agents that p is active over p being inactive. Frank $5 Eve $6 Bob -$3 Alice -$2 Dave -$5 Carol -$4 Hank $1 Gene $2
12. PARTICIPANT’S SOCIAL VALUE With Alice = 6+5-2-3 = 6 Frank $5 Eve $6 Bob -$3 Alice -$2 Dave -$5 Carol -$4 Hank $1 Gene $2 Social value for participant p how much better is for the other agents that p is active over p being inactive.
13. PARTICIPANT’S SOCIAL VALUE Without Alice = 6+5-3-4 = 4 Frank $5 Eve $6 Bob -$3 Alice -$2 Dave -$5 Carol -$4 Hank $1 Gene $2 Social value for participant p how much better is for the other agents that p is active over p being inactive.
14. PARTICIPANT’S SOCIAL VALUE Alice’s social value = 6 - 4 = 2 Social value for participant p how much better is for the other agents that p is active over p being inactive. Frank $5 Eve $6 Bob -$3 Alice -$2 Dave -$5 Carol -$4 Hank $1 Gene $2
15. CHaining Agents IN Mediated Environments CHAINME
16. CHAINME is a message passing algorithm Two phases: •Assess participants social value. •Assess the supply chain configuration.
17. Assessing participants’ Social Value
18. ASSESSING PARTICIPANTS’ SOCIAL VALUE Alice Carol Bob Dave Eve $8 $7 -$5 -$3 -$2
19. ASSESSING PARTICIPANTS’ SOCIAL VALUE Determine agent value Va = Ca + X g2Ga Sg a Alice Carol Bob Dave Eve $8 -$5 -$3 Carol Eve $8 $7 -$5 -$2 -5 -3 -2 8 7
20. ASSESSING PARTICIPANTS’ SOCIAL VALUE Determine agent value Va = Ca + X g2Ga Sg a Send offer to mediators Og a Va Sg a -2 Alice Carol Bob Dave Eve $8 -$5 -$3 Carol Eve $8 $7 -$5 -$2 -5 -3 -2 8 7 -3 -5 -5 7 8
21. ASSESSING PARTICIPANTS’ SOCIAL VALUE Determine agent value Va = Ca + X g2Ga Sg a Send offer to mediators Send updated social value Alice Carol Bob Dave Eve $8 -$5 -$3 Carol Eve $8 $7 -$5 -$2 0 0 -8 8 0 -7 Og a Va Sg a
22. ASSESSING PARTICIPANTS’ SOCIAL VALUE Determine agent value Va = Ca + X g2Ga Sg a Alice Carol Bob Dave Eve $8 -$5 -$3 Carol Eve $8 $7 -$5 -$2 3 -3 -2 1 -1
23. ASSESSING PARTICIPANTS’ SOCIAL VALUE Determine agent value Va = Ca + X g2Ga Sg a Send offer to mediators -2 Alice Carol Bob Dave Eve $8 -$5 -$3 Carol Eve $8 $7 -$5 -$2 3 -3 -2 1 -1 -3 -5 3 7 8 Og a Va Sg a
24. ASSESSING PARTICIPANTS’ SOCIAL VALUE Determine agent value Va = Ca + X g2Ga Sg a Send offer to mediators Send updated social value Alice Carol Bob Dave Eve $8 -$5 -$3 Carol Eve $8 $7 -$5 -$2 -2 3 -8 8 -1 -7 Og a Va Sg a
25. ASSESSING PARTICIPANTS’ SOCIAL VALUE Determine agent value Va = Ca + X g2Ga Sg a Alice Carol Bob Dave Eve $8 -$5 -$3 Carol Eve $8 $7 -$5 -$2 1 -2 1 1 -1
26. ASSESSING PARTICIPANTS’ SOCIAL VALUE Determine agent value Va = Ca + X g2Ga Sg a Send offer to mediators -2 Alice Carol Bob Dave Eve $8 -$5 -$3 Carol Eve $8 $7 -$5 -$2 1 -2 1 1 -1 -3 -7 3 7 8 Og a Va Sg a
27. ASSESSING PARTICIPANTS’ SOCIAL VALUE Determine agent value Va = Ca + X g2Ga Sg a Send offer to mediators Send updated social value Alice Carol Bob Dave Eve $8 -$5 -$3 Carol Eve $8 $7 -$5 -$2 -2 3 -8 8 -1 -7 Og a Va Sg a
28. Determining the supply chain configuration
29. ASSESSING PARTICIPANTS’ SOCIAL VALUE Alice Carol Bob Dave Eve $8 -$5 -$3 Carol Eve $8 $7 -$5 -$2 1 -2 1 1 -1 1. Participants with positive value anounce their availability. 2. Mediators choose active participants out of the available ones. 3. Available participants remain available if all their mediators choose them to be active. 4. Back to 2.
30. ASSESSING PARTICIPANTS’ SOCIAL VALUE Alice Carol Bob Dave Eve $8 -$5 -$3 Carol Eve $8 $7 -$5 -$2 1 -2 1 1 -1 1. Participants with positive value anounce their availability. 2. Mediators choose active participants out of the available ones. 3. Available participants remain available if all their mediators choose them to be active. 4. Back to 2.
31. Does it work?
32. SAMP-SB RB-LBP CHAINME participant memory O(G) O(G•A) O(G) mediator memory O(A) O(A) participant bandwidth O(G) O(G•A) O(G) mediator bandwidth O(A) O(A) participant operations O(G) O(G•A2) O(G) mediator operations O(logA) O(A·logA)
33. networks goods agents large 50 40-500
34. up to 50 times up to 102 times consistently close to 80% less bandwidth faster better solutions optimal solutions
35. better solutions consistently
36. optimal solutions close to 80%
37. Decentralized algorithm for SCF based on social value propagation. FINAL RECAP Reduces computation. Reduces communication. Produces (near)optimal solutions.
38. What’s next?
39. Payments Multiple Unit Time constraints Multiple Attritubes
40. Thank You
41. Questions?