Download presentation

Presentation is loading. Please wait.

Published byMaci Fiddler Modified over 2 years ago

1
Subexponential Algorithms for Unique Games and Related Problems School on Approximability, Bangalore, January 2011 David Steurer MSR New England Sanjeev Arora Princeton University & CCI Boaz Barak MSR New England & Princeton U

2
U NIQUE G AMES Goal:satisfy as many constraints as possible

3
U NIQUE G AMES Goal:satisfy as many constraints as possible Goal: Distinguish two cases Unique Games Conjecture (UGC) [Khot’02]

4
Implications of UGC For many basic optimization problems, it is NP-hard to beat current algorithms (based on simple LP or SDP relaxations) Examples: V ERTEX C OVER [Khot-Regev’03], M AX C UT [Khot-Kindler-Mossel-O’Donnell’04, Mossel-O’Donnell-Oleszkiewicz’05], every M AX C SP [Raghavendra’08], …

5
Implications of UGC For many basic optimization problems, it is NP-hard to beat current algorithms (based on simple LP or SDP relaxations) Unique Games Barrier U NIQUE G AMES is common barrier for improving current algorithms of many basic problems

6
Subexponential Algorithm for Unique Games Time vs Approximation Trade-off

7
(*) assuming 3-S AT does not have subexponential algorithms, exp( n o(1) ) Subexponential Algorithm for Unique Games Consequences

8
2-S AT 3-S AT (*) F ACTORING [Moshkovitz-Raz’08 + Håstad’97] Subexponential Algorithm for Unique Games G RAPH I SOMORPHISM

9
Subexponential Algorithm for Unique Games Interlude: Graph Expansion

10
Subexponential Algorithm for Unique Games Easy graphs for U NIQUE G AMES Constraint Graph variable vertex constraint edge Expanding constraint graph [Arora-Khot-Kolla-S.-Tulsiani-Vishnoi’08]

11
Expanding constraint graph Constraint graph with few large eigenvalues Easy graphs for U NIQUE G AMES [Arora-Khot-Kolla-S.-Tulsiani-Vishnoi’08] Subexponential Algorithm for Unique Games Graph Eigenvalues normalized adjacency matrix quasi-expander [Kolla-Tulsiani’07, Kolla’10, here, Barak-Raghavendra-S.’11]

12
Expanding constraint graph Constraint graph with few large eigenvalues Easy graphs for U NIQUE G AMES [Arora-Khot-Kolla-S.-Tulsiani-Vishnoi’08] [Kolla-Tulsiani’07, Kolla’10, here, Barak-Raghavendra-S.’11] Subspace Enumeration [Kolla-Tulsiani’07] exhaustive search over certain subspace Subexponential Algorithm for Unique Games Question: Dimension of this subspace? quasi-expander

13
[Kolla-Tulsiani’07, Kolla’10, here, Barak-Raghavendra-S.’11] quasi-expander Expanding constraint graph Constraint graph with few large eigenvalues Easy graphs for U NIQUE G AMES [Arora-Khot-Kolla-S.-Tulsiani-Vishnoi’08] Subexponential Algorithm for Unique Games Graph Decomposition Approach [Trevisan’05, Arora-Impagliazzo-Matthews-S.’10] hard constraint graph remove 1% of edges easy

14
Approach [Trevisan’05, Arora-Impagliazzo-Matthews-S.’10] quasi-expander Expanding constraint graph Constraint graph with few large eigenvalues Easy graphs for U NIQUE G AMES [Arora-Khot-Kolla-S.-Tulsiani-Vishnoi’08] [Kolla-Tulsiani’07, Kolla’10, here] Subexponential Algorithm for Unique Games Graph Decomposition

15
Approach [Trevisan’05, Arora-Impagliazzo-Matthews-S.’10] Expanding constraint graph Constraint graph with few large eigenvalues Easy graphs for U NIQUE G AMES [Arora-Khot-Kolla-S.-Tulsiani-Vishnoi’08] [Kolla-Tulsiani’07, Kolla’10, here] Subexponential Algorithm for Unique Games Graph Decomposition ? YES ?

16
Approach [Trevisan’05, Arora-Impagliazzo-Matthews-S.’10] Expanding constraint graph Constraint graph with few large eigenvalues Easy graphs for U NIQUE G AMES [Arora-Khot-Kolla-S.-Tulsiani-Vishnoi’08] [Kolla-Tulsiani’07, Kolla’10, here] Subexponential Algorithm for Unique Games Graph Decomposition

17
Approach [Trevisan’05, Arora-Impagliazzo-Matthews-S.’10] Expanding constraint graph Constraint graph with few large eigenvalues Easy graphs for U NIQUE G AMES [Arora-Khot-Kolla-S.-Tulsiani-Vishnoi’08] [Kolla-Tulsiani’07, Kolla’10, here] Subexponential Algorithm for Unique Games Graph Decomposition NO ? ?

18
[Kolla-Tulsiani’07, Kolla’10, here, Barak-Raghavendra-S.’11] quasi-expander Expanding constraint graph Constraint graph with few large eigenvalues Easy graphs for U NIQUE G AMES [Arora-Khot-Kolla-S.-Tulsiani-Vishnoi’08] Subexponential Algorithm for Unique Games Graph Decomposition By removing 1% of edges, every graph can be decomposed into components with Classical Here [Leighton-Rao’88, Goldreich-Ron’98, Spielman-Teng’04, Trevisan’05]

19
Expanding constraint graph Constraint graph with few large eigenvalues Easy graphs for U NIQUE G AMES [Arora-Khot-Kolla-S.-Tulsiani-Vishnoi’08] Subexponential Algorithm for Unique Games Graph Decomposition By removing 1% of edges, every graph can be decomposed into components with Classical Here quasi-expander [Kolla-Tulsiani’07, Kolla’10, here, Barak-Raghavendra-S.’11]

20
Constraint graph with few large eigenvalues [Kolla-Tulsiani’07, Kolla’10, here, Barak-Raghavendra-S.’11] 2

21
Constraint graph with few large eigenvalues [Kolla-Tulsiani’07, Kolla’10, here, Barak-Raghavendra-S.’11] 99% of indicator vector lies in span of top m eigenvectors enumerate subspace 2

22
Constraint graph with few large eigenvalues [Kolla-Tulsiani’07, Kolla’10, here, Barak-Raghavendra-S.’11] 99% of indicator vector lies in span of top m eigenvectors enumerate subspace Compare: Fourier-based learning 2

23
Constraint graph with few large eigenvalues [Kolla-Tulsiani’07, Kolla’10, here, Barak-Raghavendra-S.’11]

24
Easy graphs for U NIQUE G AMES Constraint graph with few large eigenvalues [Kolla-Tulsiani’07, Kolla’10, here, Barak-Raghavendra-S.’11] Graph Decomposition By removing 1% of edges, every graph can be decomposed into components with Classical Here Subexponential Algorithm for Unique Games

25
Constraint graph with few large eigenvalues [Kolla-Tulsiani’07, Kolla’10, here, Barak-Raghavendra-S.’11] Easy graphs for U NIQUE G AMES Graph Decomposition By removing 1% of edges, every graph can be decomposed into components with Classical Here Subexponential Algorithm for Unique Games

26
Graph Decomposition By removing 1% of edges, every graph can be decomposed into components with follows from Cheeger bound [Cheeger’70, Alon-Milman’85, Alon’86] V S Assume graph is d-regular

27
Graph Decomposition By removing 1% of edges, every graph can be decomposed into components with Assume graph is d-regular follows from Cheeger bound V S “higher-order Cheeger bound” [here]

28
“higher-order Cheeger bound”

29

30
Idea

31
“higher-order Cheeger bound” It would suffice to show:

32
“higher-order Cheeger bound” It would suffice to show: collision probability of t-step random walk from i Suffices local Cheeger bound

33
“higher-order Cheeger bound” It would suffice to show: “A Markov chain with many large eigenvalues cannot mix locally everywhere” collision probability of t-step random walk from i Suffices

34
Improved approximations for d- TO -1 G AMES ( Khot’s d-to-1 Conjecture) [S.’10] Better approximations for M AX C UT and V ERTEX C OVER on small-set expanders Open Questions How many large eigenvalues can a small-set expander have? Thank you!Questions? More Subexponential Algorithms Similar approximation for M ULTI C UT and S MALL S ET E XPANSION What else can be done in subexponential time? Towards refuting the Unique Games Conjecture Better approximations for M AX C UT or V ERTEX C OVER on general instances?

Similar presentations

© 2017 SlidePlayer.com Inc.

All rights reserved.

Ads by Google