Differentially Private and Strategy Proof Spectrum Auction with

  • Slides: 43
Download presentation
Differentially Private and Strategy. Proof Spectrum Auction with Approximate Revenue Maximization Ruihao Zhu and

Differentially Private and Strategy. Proof Spectrum Auction with Approximate Revenue Maximization Ruihao Zhu and Kang G. Shin Department of Electrical Engineering and Computer Science, University of Michigan, Ann Arbor 1

Outline Ø Background Ø Design Goal Ø Primers: differential privacy, exponential mech. , truthfulness

Outline Ø Background Ø Design Goal Ø Primers: differential privacy, exponential mech. , truthfulness , revenue maximization Ø Near Optimal Mechanism Ø PASS Ø Evaluation Results 22

Spectrum Need Forecast ‐ Table of Results FCC whitepaper, Oct. 2010 3

Spectrum Need Forecast ‐ Table of Results FCC whitepaper, Oct. 2010 3

Secondary Spectrum Market Traditionally, static, long-term licenses – Radio spectrum is not fully utilized

Secondary Spectrum Market Traditionally, static, long-term licenses – Radio spectrum is not fully utilized – Unlicensed bands are getting crowded =>Dynamic spectrum redistribution/auction needed! 4

Unique Challenge in Spectrum Auctions • Spatial Reusability – Bidders far away can use

Unique Challenge in Spectrum Auctions • Spatial Reusability – Bidders far away can use the same channel Channel 1 Channel 2 5

Traditional Spectrum Auctions Auctioneer Bidders Channels Auctioneer’s Revenue Truthfulness 6

Traditional Spectrum Auctions Auctioneer Bidders Channels Auctioneer’s Revenue Truthfulness 6

Privacy in Spectrum Auctions • Channels are for short-term usage. • Sequential auctions make

Privacy in Spectrum Auctions • Channels are for short-term usage. • Sequential auctions make inference of bidding information possible even with secure channel. 7

Privacy in Spectrum Auctions How to infer? 8

Privacy in Spectrum Auctions How to infer? 8

Privacy in Spectrum Auctions, cont’d 0. 01% revenue for channel cost 9

Privacy in Spectrum Auctions, cont’d 0. 01% revenue for channel cost 9

Outline Ø Background Ø Design Goal Ø Primers: differential privacy, exponential mech. , truthfulness

Outline Ø Background Ø Design Goal Ø Primers: differential privacy, exponential mech. , truthfulness Ø Near Optimal Mechanism Ø PASS Ø Evaluation Results 10

Goal Design a truthful auction mechanism that maximizes auctioneer’s revenue while keeping participants’ bidding

Goal Design a truthful auction mechanism that maximizes auctioneer’s revenue while keeping participants’ bidding prices confidential 11

Outline Ø Background Ø Design Goal Ø Primers: differential privacy, exponential mech. , truthfulness,

Outline Ø Background Ø Design Goal Ø Primers: differential privacy, exponential mech. , truthfulness, revenue maximization Ø Near Optimal Mechanism Ø PASS Ø Evaluation Results 12

Differential Privacy 13

Differential Privacy 13

Differential Privacy, cont’d • 14

Differential Privacy, cont’d • 14

Differential Privacy cont’d • Randomness (no deterministic DP): - Input perturbation - Exponential mechanism

Differential Privacy cont’d • Randomness (no deterministic DP): - Input perturbation - Exponential mechanism 15

Exponential Mechanism • 16

Exponential Mechanism • 16

Truthful (in Expectation) 17

Truthful (in Expectation) 17

Truthful Mechanism 18

Truthful Mechanism 18

Revenue Maximization 19

Revenue Maximization 19

Outline Ø Background Ø Problem Definition Ø Primers: differential privacy, exponential mech. , truthfulness,

Outline Ø Background Ø Problem Definition Ø Primers: differential privacy, exponential mech. , truthfulness, revenue maximization Ø Near Optimal Mechanism Ø PASS Ø Evaluation Results 20

Near Optimal Mechanism • 21

Near Optimal Mechanism • 21

Outline Ø Background Ø Design Goal Ø Primers: differential privacy, exponential mech. , truthfulness

Outline Ø Background Ø Design Goal Ø Primers: differential privacy, exponential mech. , truthfulness , revenue maximization Ø Near Optimal Mechanism Ø PASS Ø Evaluation Results 22

Illustrative Example location 4 2 5 Interference range 1 3 23

Illustrative Example location 4 2 5 Interference range 1 3 23

PASS Graph Partition Virtual Channel Random Selection and Allocation 24

PASS Graph Partition Virtual Channel Random Selection and Allocation 24

PASS Graph Partition entire area uniformly into small hexagons with side length equal half

PASS Graph Partition entire area uniformly into small hexagons with side length equal half interference range. 4 5 2 1 3 25

PASS Virtual Channel 4 5 2 1 3 26

PASS Virtual Channel 4 5 2 1 3 26

PASS Random Selection and Allocation 4 5 2 1 3 27

PASS Random Selection and Allocation 4 5 2 1 3 27

PASS Random Selection and Allocation 4 5 2 1 3 28

PASS Random Selection and Allocation 4 5 2 1 3 28

PASS Random Selection and Allocation Suppose bidder 1 is selected. 4 5 2 1

PASS Random Selection and Allocation Suppose bidder 1 is selected. 4 5 2 1 3 29

PASS Random Selection and Allocation All the bidders conflict with bidder 1 is removed.

PASS Random Selection and Allocation All the bidders conflict with bidder 1 is removed. 4 5 2 1 3 30

PASS Random Selection and Allocation 4 5 31

PASS Random Selection and Allocation 4 5 31

PASS Random Selection and Allocation Suppose bidder 5 is selected. 4 5 32

PASS Random Selection and Allocation Suppose bidder 5 is selected. 4 5 32

PASS Random Selection and Allocation All the bidders conflict with bidder 5 is removed.

PASS Random Selection and Allocation All the bidders conflict with bidder 5 is removed. 4 5 33

Properties of PASS 34

Properties of PASS 34

Outline Ø Background Ø Design Goal Ø Primers: differential privacy, exponential mech. , truthfulness

Outline Ø Background Ø Design Goal Ø Primers: differential privacy, exponential mech. , truthfulness , revenue maximization Ø Near Optimal Mechanism Ø PASS Ø Evaluation Results 35

Revenue of PASS (5 channels) 36

Revenue of PASS (5 channels) 36

Revenue of PASS (10 channels) 37

Revenue of PASS (10 channels) 37

Revenue of PASS (15 channels) 38

Revenue of PASS (15 channels) 38

 Privacy of PASS (5 channels) 39

Privacy of PASS (5 channels) 39

 Privacy of PASS (10 channels) 40

Privacy of PASS (10 channels) 40

 Privacy of PASS (15 channels) 41

Privacy of PASS (15 channels) 41

Conclusion • PASS: First differentially private and truthful spectrum auction mechanism with approximate revenue

Conclusion • PASS: First differentially private and truthful spectrum auction mechanism with approximate revenue maximization. • Theoretically proved the properties in revenue and privacy. • Implemented PASS and extensively evaluated its performance. 42

Thank you! rhzhu@umich. edu 43

Thank you! rhzhu@umich. edu 43