Download presentation

Presentation is loading. Please wait.

Published byVeronica Nelson Modified over 3 years ago

1
Recent Progress in Approximability

2
Administrivia Most agreeable times: Monday 2:30-4:00 Wednesday 4:00-5:30 Thursday 4:00-5:30 Friday 1:00-2:30 Please Fill Up Survey: http://www.surveymonkey.com/s/9TSVQM7 http://www.surveymonkey.com/s/9TSVQM7 Evaluation: 6-8 short homeworks and class participation.

3
Max Cut 10 15 3 7 1 1 Max CUT Input: A weighted graph G Find: A Cut with maximum number/weight of crossing edges Fraction of crossing edges MaxCut is NP-complete (Karp’s original list of 21 NP- complete problems (1971)

4
An algorithm A is an α -approximation for a problem if for every instance I, A (I) ≥ α ∙ OPT(I) --Vast Literature-- Approximation Algorithms

5
Max Cut 10 15 3 7 1 1 Max CUT Input: A weighted graph G Find: A Cut with maximum number/weight of crossing edges Trivial ½ Approximation Assign each vertex randomly to left or right side of the cut Analysis For every edge e, Probability[edge is cut] = ½ Fraction of edges cut = ½ Optimum MaxCut < 1 So, Solution returned = ½ > ½ *Optimum MaxCut Till 1994, this was the state of the art. Many linear programming techniques were known to NOT get any better approximation.

6
The Tools Till 1994, A majority of approximation algorithms directly or indirectly relied on Linear Programming. In 1994, Semidefinite Programming based algorithm for Max Cut [Goemans-Williamson] Semidefinite Programming - A generalization of Linear Programming. Semidefinite Programming is the one of the most powerful tools in approximation algorithms.

7
Semidefinite Program Variables : v 1, v 2 … v n | v i | 2 = 1 Maximize Max Cut SDP Quadratic Program Variables : x 1, x 2 … x n x i = 1 or -1 Maximize 10 15 3 7 1 1 1 1 1 Relax all the x i to be unit vectors instead of {1,-1}. All products are replaced by inner products of vectors 1

8
Semidefinite Program: [Goemans-Williamson 94] Embedd the graph on the N - dimensional unit ball, Maximizing ¼ ( Average Squared Length of the edges ) Semidefinite Program [Goemans-Williamson 94] Variables : v 1, v 2 … v n |v i | 2 = 1 Maximize MaxCut 10 15 3 7 1 1 1 1 1 1 Max Cut Problem Given a graph G, Find a cut that maximizes the number of crossing edges v1v1 v2v2 v3v3 v4v4 v5v5

9
MaxCut Rounding v1v1 v2v2 v3v3 v4v4 v5v5 Cut the sphere by a random hyperplane, and output the induced graph cut. -A 0.878 approximation for the problem. [Goemans-Williamson]

10
Analysis v1v1 v2v2 v3v3 v4v4 v5v5 SDP Optiumum 10 15 3 7 1 1 Optimal MaxCut v1v1 v2v2 v3v3 v4v4 v5v5 Algorithm’s Output 01 Rounding Ratio > 0.878 Integrality Gap Algorithm Output > 0.878 X SDP Optimum > 0.878 X Optimum MaxCut

11
minimum over all instances = value of rounded solution value of SDP solution rounding – ratio A (approximation ratio) ≤ integrality gap = value of optimal solution value of SDP solution minimum over all instances For any rounding algorithm A, and a SDP relaxation ¦ v1v1 v2v2 v3v3 v4v4 v5v5 SDP Optiumum 10 15 3 7 1 1 Optimal MaxCut v1v1 v2v2 v3v3 v4v4 v5v5 Algorithm’s Output 01 Rounding Ratio > 0.878 Integrality Gap = “algorithm achieves the gap’’

12
Inapproximability Is 0.878 the best possible approximation ratio for MaxCut? Satisfiable Unsatisfiable MaxCut value = K MaxCut value < K 10 15 3 7 1 1 1 1 1 3-SAT Instance Polynomial time reduction

13
What we need.. (Completeness) Satisfiable (Soundness) Unsatisfiable MaxCut value = K MaxCut value < 0.9K 10 15 3 7 1 1 1 1 1 3-SAT Instance Polynomial time reduction If we had a polytime 0.95 approximation algorithm for MaxCut A polytime algorithm for 3-SAT

14
A probabilistically checkable proof (PCP) Goal: Alice wants to prove to Bob that 3-SAT instance A is satisfiable 3-SAT Instance A Alex Bob (polytime machine) Satisfying assignment

15
A probabilistically checkable proof (PCP) Goal: Alice wants to prove to Bob that 3-SAT instance A is satisfiable 3-SAT Instance A Alex Bob (polytime machine) 10 15 3 7 1 1 1 1 1 Polynomial time reduction 3-SAT Instance A 10 15 3 7 1 1 1 1 1 Polynomial time reduction Probabilistically Checkable Proof A cut of value > 0.9 Verifier (Bob): Sample a random edge in graph, Accept if edge is cut. Prob[Bob Accepts] = Value of the Cut

16
Suppose, (Completeness) Satisfiable (Soundness) Unsatisfiable MaxCut value = 0.99 MaxCut value < 0.9 10 15 3 7 1 1 1 1 1 3-SAT Instance Polynomial time reduction Completeness: There exists a ``proof” that Bob accepts with probability 0.99 Soundness: No matter what Alex does, Bob accepts with probability < 0.9 Bob reads only 2 bits of the proof!!

17
Analogy to Math Proofs Could you check the proof of a theorem with any reasonable confidence by reading only 3 bits of the proof??? Guess: Probably Not.. Max-SNP complexity class was defined, because it was believable that we will never be able to get a Gap Reduction aka Probabilistically Checkable Proof for NP.

18
PCP Theorem: [Arora-Lund-Motwani-Sudan-Szegedy 1991] Max-3-SAT is NP-hard to approximate better than 1- 10^{-100}. Corollary: Max-Cut is NP-hard to approximate better than 1- 10^{-200}. Long and very difficult proof, simplified over the years.. (*Check out History of PCP Theorem: http://www.cs.washington.edu/education/courses/cse533/05au/pcp-history.pdf) Completely new proof by Irit Dinur in 2005.

19
Hastad’s 3-Query PCP [ Håstad STOC97] For any ε > 0, NP has a 3-query probabilistically checkable proof system such that: Completeness = (1 – ε) Soundness = 1/2 + ε Verifier reads only 3-bits, and checks a linear equation on them! X i + X j = X k + c (mod p) Alternately,

20
Hastad’s 3-Query PCP [1997] For any ε > 0, given a set of linear equations modulo 2, it is NP-hard to distinguish between: (1 – ε) – fraction of the equations can be satisfied. 1/2 + ε – fraction of the equations can be satisfied. All equations are of the form X i + X j = X k + c (mod p) By Very Clever Gadget reductions, [Sudan-Sorkin-Trevisan-Williamson] MaxCut is NP-hard to approximate beyond 0.94.

21
ALGORITHMS [Charikar-Makarychev-Makarychev 06] [Goemans-Williamson] [Charikar-Wirth] [Lewin-Livnat-Zwick] [Charikar-Makarychev-Makarychev 07] [Hast] [Charikar-Makarychev-Makarychev 07] [Frieze-Jerrum] [Karloff-Zwick] [Zwick SODA 98] [Zwick STOC 98] [Zwick 99] [Halperin-Zwick 01] [Goemans-Williamson 01] [Goemans 01] [Feige-Goemans] [Matuura-Matsui] [Trevisan-Sudan-Sorkin-Williamson] Approximability of CSPs Gap for MaxCUT Algorithm = 0.878 Hardness = 0.941 MAX CUT MAX 2-SAT MAX 3-SAT MAX 4-SAT MAX DI CUT MAX k-CUT Unique Games MAX k-CSP MAX Horn SAT MAX 3 DI-CUT MAX E2 LIN3 MAX 3-MAJ MAX 3-CSP MAX 3-AND 01 NP HARD

22
Given linear equations of the form: X i – X k = c ik mod p Satisfy maximum number of equations. x-y = 11 (mod 17) x-z = 13 (mod 17) … …. z-w = 15(mod 17) Unique Games Conjecture [Khot 02] [KKMO] For every ε> 0, for large enough p, Given : 1-ε (99%) satisfiable system, NP-hard to satisfy ε (1%) fraction of equations. Towards bridging this gap, In 2002, Subhash Khot introduced the Unique Games Conjecture

23
A notorious open problem. Hardness Results: No constant factor approximation for unique games. [Feige- Reichman] Algorithm On (1-Є) satisfiable instances [Khot 02] [Trevisan] [Gupta-Talwar] 1 – O(ε logn) [Charikar-Makarychev-Makarychev] [Chlamtac-Makarychev-Makarychev] [Arora-Khot-Kolla-Steurer-Tulsiani-Vishnoi]

24
Assuming UGC UGC Hardness Results [Khot-Kindler-Mossel-O’donnell] [Austrin 06] [Austrin 07] [Khot-Odonnell] [Odonnell-Wu] [Samorodnitsky-Trevisan] NP HARDUGC HARD 01 MAX CUT MAX 2-SAT MAX 3-SAT MAX 4-SAT MAX DI CUT MAX k-CUT Unique Games MAX k-CSP MAX Horn SAT MAX 3 DI-CUT MAX E2 LIN3 MAX 3-MAJ MAX 3-CSP MAX 3-AND For MaxCut, Max-2-SAT, Unique Games based hardness = approximation obtained by Semidefinite programming!

25
The Connection MAX CUT MAX 2-SAT MAX 3-SAT MAX 4-SAT MAX DI CUT MAX k-CUT Unique Games MAX k-CSP MAX Horn SAT MAX 3 DI-CUT MAX E2 LIN3 MAX 3-MAJ MAX 3-CSP MAX 3-AND 01 UGC Hard GENERIC ALGORITHM Theorem: Assuming Unique Games Conjecture, For every CSP, “the simplest semidefinite programs give the best approximation computable efficiently.”

26
Constraint Satisfaction Problems [Raghavendra`08][Austrin-Mossel] M AX C UT [Khot-Kindler-Mossel-ODonnell][Odonnell-Wu] M AX 2S AT [Austrin07][Austrin08] Ordering CSPs [Charikar-Guruswami-Manokaran-Raghavendra-Hastad`08] M AX A CYCLIC S UBGRAPH, B ETWEENESS Grothendieck Problems [Khot-Naor, Raghavendra-Steurer] Metric Labeling Problems [Manokaran-Naor-Raghavendra-Schwartz`08] M ULTIWAY C UT, 0- EXTENSION Kernel Clustering Problems [Khot-Naor`08,10] Strict Monotone CSPs [Kumar-Manokaran-Tulsiani-Vishnoi`10] V ERTEX C OVER [Khot-Regev], H YPERGRAPH V ERTEX C OVER Assuming the Unique Games Conjecture, A simple semidefinite program (Basic-SDP) yields the optimal approximation ratio for Is the conjecture true? Many many ways to disprove the conjecture! Find a better algorithm for any one of these problems.

27
The UG Barrier Constraint Satisfaction Problems Graph Labelling Problems Ordering CSPs Kernel Clustering Problems Monotone Min-One CSPs UGC HARD If UGC is true, Then Simplest SDPs give the best approximation possible. If UGC is false, Hopefully, a new algorithmic technique will arise.

28
What if UGC is false? Could existing techniques ( LPs/SDPs) disprove the UGC?

29
What if UGC is false?

30
UGC is false New algorithms? Unique Games Constraint Satisfaction Problems [Raghavendra`08] M AX C UT, M AX 2S AT Ordering CSPs [GMR`08] M AX A CYCLIC S UBGRAPH, B ETWEENESS Grothendieck Problems [KNS`08, RS`09] Metric Labeling Problems [MNRS`08] M ULTIWAY C UT, 0- EXTENSION Kernel Clustering Problems [KN`08,10] Strict Monotone CSPs [KMTV`10] V ERTEX C OVER, H YPERGRAPH V ERTEX C OVER … Problem X UGC is false New algorithm for Problem X Despite considerable efforts, No such reverse reduction known for any of the above problems [Feige-Kindler-Odonnell,Raz’08, BHHRRS’08]

31
Graph Expansion d-regular graph G d expansion(S) = # edges leaving S d |S| vertex set S A random neighbor of a random vertex in S is outside of S with probability expansion(S) Ф G = expansion(S) minimum |S| ≤ n/2 Conductance of Graph G Uniform Sparsest Cut Problem Given a graph G compute Ф G and find the set S achieving it. Approximation Algorithms: Cheeger’s Inequality [Alon][Alon-Milman] Given a graph G, if the normalized adjacency matrix has second eigen value λ 2 then, A log n approximation algorithm [Leighton-Rao 98-99?]. A sqrt(log n) approximation algorithm using semidefinite programming [Arora-Rao-Vazirani 2004]. Extremely well-studied, many different contexts pseudo-randomness, group theory, online routing, Markov chains, metric embeddings, …

32
A Reverse Reduction Graph (Social Network) Close-knit community Finding Small Non Expanding Sets Suppose there exists is a small community say (0.1% of the population) 99% of whose friends are within the community.. Find one such close-knit community. Theorem [R-Steurer 10] UGC is false New algorithms to approximate expansion of small sets in graphs

33
STILL OPEN: Reverse reduction from Max Cut or Vertex Cover to Unique Games.

34
What if UGC is false? Could existing algorithmic techniques (LPs/SDPs) disprove the UGC?

35
Could LPs/SDPs disprove the UGC?

36
Question I: Could some small L INEAR P ROGRAM give a better approximation for MaxCut or Vertex Cover thereby disproving the UGC? Probably Not! [Charikar-Makarychev-Makarychev][Schoenebeck-Tulsiani] For MaxCut, for several classes of linear programs, exponential sized linear programs are necessary to even beat the trivial ½ approximation! Question II: Could some small S EMI D EFINITE P ROGRAM give a better approximation for MaxCut or Vertex Cover thereby disproving the UGC? We don’t know.

37
v1v1 v2v2 v3v3 v4v4 v5v5 Max Cut SDP: Embedd the graph on the N - dimensional unit ball, Maximizing ¼ ( Average squared length of the edges ) In the integral solution, all the vectors v i are 1,-1. Thus they satisfy additional constraints For example : (v i – v j ) 2 + (v j – v k ) 2 ≥ (v i – v k ) 2 (the triangle inequality) The Simplest Relaxation for MaxCut Does adding triangle inequalities improve approximation ratio? (and thereby disprove UGC!)

39
Until 2009: Adding a simple constraint on every 5 vectors could yield a better approximation for MaxCut,and disproves UGC! Building on the work of [Khot-Vishnoi],

41
Constraint Satisfaction Problems Max 3 SAT Find an assignment that satisfies the maximum number of clauses. Variables Finite Domain Constraints {x 1,x 2, x 3, x 4, x 5 } {0,1} Clauses Kind of constraints permitted Different CSPs

42
Deeper understanding of the UGC – why it should be true if it is. Why play this game? Connections between SDP hierarchies, Spectral Graph Theory and Graph Expansion. New algorithms based on SDP hierarchies. [Raghavendra-Tan] Improved approximation for MaxBisection using SDP hierarchies [Barak-Raghavendra-Steurer] Algorithms for 2-CSPs on low-rank graphs. New Gadgets for Hardness Reductions: [Barak-Gopalan-Hastad-Meka-Raghavendra-Steurer] A more efficient long code gadget.

Similar presentations

OK

Lecture 22: April 18 Probabilistic Method. Why Randomness? Probabilistic method: Proving the existence of an object satisfying certain properties without.

Lecture 22: April 18 Probabilistic Method. Why Randomness? Probabilistic method: Proving the existence of an object satisfying certain properties without.

© 2018 SlidePlayer.com Inc.

All rights reserved.

Ads by Google

Ppt on biodegradable and non biodegradable bins Ppt on video teleconferencing certifications Ppt on amplitude shift keying modem Ppt on electricity for class 10th exam Ppt on types of pollution download Ppt on formal education definition Ppt on pirates of silicon valley Ppt on pricing policy in international marketing Ppt on db2 introduction to accounting Ppt on model view controller definition