1 / 32

How to Round Any CSP

How to Round Any CSP. (In Principle). Prasad Raghavendra University of Washington, Seattle David Steurer , Princeton University. Constraint Satisfaction Problem A Classic Example : Max-3-SAT. Given a 3-SAT formula,

avi
Download Presentation

How to Round Any CSP

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. How to Round Any CSP (In Principle) Prasad Raghavendra University of Washington, Seattle David Steurer, Princeton University

  2. Constraint Satisfaction ProblemA Classic Example : Max-3-SAT Given a 3-SAT formula, Find an assignment to the variables that satisfies the maximum number of clauses. Equivalently the largest fraction of clauses

  3. Constraint Satisfaction Problem Instance : • Set of variables. • Predicates Pi applied on variables Find an assignment that satisfies the largest fraction of constraints. Problem : Domain : {0,1,.. q-1} Predicates : {P1, P2 , P3 … Pr} Pi : [q]k -> {0,1} Max-3-SAT Domain : {0,1} Predicates : P1(x,y,z) = x ѵ y ѵz Variables : {x1 , x2 , x3 ,x4 , x5} Constraints : 4 clauses

  4. Theorem: [Raghavendra 08] Assuming Unique Games Conjecture, For every CSP, “a simple semidefinite program (SDP1) gives the best approximation computable efficiently.” [Raghavendra08] A generic rounding scheme for (SDP1) that is optimal for every CSP under UGC. Independent of UGC, for 2CSPs, the generic rounding scheme for (SDP1) achieves an Approximation Ratio ≥ (1-²) Integrality Gap of SDP.

  5. Rounding Algorithm For any CSP ¦and any ²>0, there exists an efficient algorithm A, value of optimal solution value of SDP solution minimum over all instances = rounding – ratioA ( ¦ ) (approximation ratio) ≥ (1-²) integrality gap ( ¦ ) • Drawbacks • Running Time(A) • On CSP over alphabet size q, arity k • No explicit approximation ratio Unconditionally, the algorithm A as good as all known algorithms for CSPs Very Simple : No Invariance Principle, Dictatorship Tests, Unique Games. minimum over all instances value of rounded solution value of SDP solution =

  6. Computing Integrality Gaps Theorem: For any CSP ¦and any ²>0, there exists an algorithm A to compute integrality gap (¦) within an accuracy ² Running Time(A) On CSP over alphabet size q, arity k

  7. SDP ALGORITHMS [Charikar-Makarychev-Makarychev 06] MaxCut [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] Previous Work [O’Donnell-Wu] Optimal rounding schemes for MaxCut

  8. Rounding Any Constraint Satisfaction Problem AlGORITHM OUTLINE

  9. Max Cut Max CUT Input : A weighted graph G Find : A cut with maximum fraction of crossing edges 10 15 7 1 1 3

  10. Max Cut SDP Semidefinite Program Variables : v1 , v2 … vn • | vi |2= 1 Maximize -1 10 1 -1 15 1 7 1 1 1 -1 -1 -1 3 -1

  11. v2 MaxCutRounding Problem Given a graph on the n - dimensional unit ball, Find the maximum cut of the graph. v1 v3 v5 v4

  12. Approximation using Finite Models 1 -1 10 approximate solution for = ¦-CSP Instance = 1 -1 15 7 -1 1 1 1 -1 1 variable folding unfolding of the assignment -1 (identifying variables) 3 -1 1 optimal solution for =finite ¦-CSP Instance =finite 1 constant time -1 Challenge: ensure =finite has a good solution

  13. Approximation using Finite Models PTAS for dense instances General Method for CSPs [Frieze-Kannan] For a dense instance =, it is possible to construct finite model =finite OPT(=finite) ≥ (1-ε) OPT(=) What we will do : SDP value (=finite) > (1-ε)SDP value (=)

  14. Analysis of Rounding Scheme ¦-CSP Instance = ¦-CSP Instance =finite SDP value > ® - ² SDP value ® unfolding 010001001 rounded value ¯ 010001001 OPT value ¯ Hence: rounding-ratio for = < (1+²) integrality-ratio for =finite

  15. Rounding Any Constraint Satisfaction Problem Constructing FINITE MODELS (MAXCUT)

  16. STEP 1 : Dimension Reduction • Pick d = 1/ Є4 random Gaussian vectors {G1 , G2 , .. Gd} • Project the SDP solution along these directions. • Map vector V • V → V’ = (V∙G1 , V∙G2 , … V∙Gd) v2 v1 v3 v2 STEP 2 : Surgery Scale every vector V’ to unit length v2 v1 v3 • STEP 3 : Discretization • Pick an Є–netfor the • d dimensional sphere • Move every vertex to the nearest point in the Є–net v4 v5 v5 Constant dimensions v4 FINITE MODEL Graph on Є–net points

  17. To Show:SDP value (=finite) > (1-ε)SDP value (=) Lemma : “Inner Products are almost preserved under random projections” If V’,U’ are random projections of U, V on 1/ε4 directions, Pr [ |V∙U – V’∙U’| > ε] < ε2

  18. To Show:SDP value (=finite) > (1-ε)SDP value (=) For SDP value (=) Contribution of an edge e = (U,V) |U-V|2 = 2-2 V∙U SDP Vectors for =finite = Corresponding vectors in Є–net • STEP 1 : Dimension Reduction • Project the SDP solution along 1/ Є4 random directions. STEP 1 With probability > 1- Є2 , ||U-V|2 - |U’-V’|2 | < 2Є STEP 2 With probability > 1- 2Є2 , 1+ Є< |V’|2 ,|U’|2 < 1- Є, Normalization changes distance by at most 2Є STEP 2 : Surgery Scale every vector V’ to unit length • STEP 3 : Discretization • Pick an Є–netfor the • d dimensional sphere • Move every vertex to the nearest point in the Є–net STEP 3 Changes edge length by at most 2Є

  19. To Show:SDP value (=finite) > (1-ε)SDP value (=) For SDP value (=) Contribution of an edge e = (U,V) |U-V|2 = 2-2 V∙U SDP Vectors for =finite = Corresponding vectors in Є–net ANALYSIS With probability 1-3Є2, The contribution of edge e changes by < 6Є In expectation, For (1-3Є2) edges, the contribution of edge e changes by < 6Є SDP value (=finite) > SDP value (=) - 6Є – 3Є2 STEP 1 With probability > 1- Є2 , ||U-V|2 - |U’-V’|2 | < 2Є STEP 2 With probability > 1- 2Є2 , 1+ Є< |V’|2 ,|U’|2 < 1- Є, Normalization changes distance by at most 2Є STEP 3 Changes edge length by at most 2Є

  20. Rounding Any Constraint Satisfaction Problem FINITE MODELS FOR GENERAL CSP

  21. SemidefiniteProgram for CSPs Constraints : For each clause P, 0 ≤μ(P,α)≤ 1 For each clause P (xaνxbνxc), For each pair Xa,Xb in P, consitency between vector and LP variables. V(a,0) ∙V(b,0) = μ(P,000)+ μ(P,001) V(a,0) ∙V(b,1) = μ(P,010) + μ(P,011) V(a,1)∙V(b,0) = μ(P,100) + μ(P,101) V(a,1)∙V(b,1) = μ(P,100) + μ(P,101) Variables : For each variable Xa Vectors {V(a,0) , V(a,1)} For each clause P = (xaνxbνxc), Scalar variables μ(P,000) , μ(P,001) , μ(P,010) , μ(P,100) , μ(P,011) , μ(P,110) , μ(P,101) , μ(P,111) Objective Function : Xa = 1 V(a,0) = 0 V(a,1) = 1 Xa = 0 V(a,0) = 1 V(a,1) = 0 If Xa = 0, Xb = 1, Xc = 1 μ(P,000) = 0 μ(P,011) = 1 μ(P,001) = 0 μ(P,110) = 0 μ(P,010) = 0 μ(P,101) = 0 μ(P,100) = 0 μ(P,111) = 0

  22. Semidefinite Relaxation for CSP SDP solution for =: Example of local distr.: Á = 3XOR(x3, x4, x7) • for every constraint Á in = • local distributions ¹Á over assignments to the variables of Á x3x4x7 ¹Á 0 0 0 0.1 0 0 1 0.01 0 1 0 0 … 1 1 1 0.6 • for every variable xi in = • vectors vi,1 , … , vi,q Explanation of constraints: first and second moments of distributions are consistent and form PSD matrix constraints (also for first moments) SDP objective: maximize

  23. Strong and Weak STRENGTH For every clause Á in = • local distributions ¹Á over assignments to the variables of Á Vector variables vi,a within a clause Á satisfy all valid constraints (like triangle inequality) – the inner products are in the integral hull. WEAKNESS • The above hard constraint is only for variables that participate together in a clause

  24. Throwing away constraints Throw away constraints from SDP relaxation {vi,a } { μ …} -Infeasible SDP solution for a instance = , it does not satisfy the consistency for a clause P. Consider instance =‘ = = - {P} Now {vi,a } { μ … } is a good SDP solution for =‘ Throw away clauses from CSP

  25. STEP 1 : Dimension Reduction • Project the SDP solution along • d =1/ Є4 random directions. v2 STEP 2 : Throw away Discard clauses for which the corresponding inner products are not preserved within Є. =‘ = New instance v1 v3 v2 v2 v1 • STEP 3 : Discretization • Pick an Є–netfor the • d dimensional sphere • Move every variable to the nearest point in the Є–net • =finite = discretized instance v3 v4 v5 v5 Constant dimensions v4 FINITE MODEL CSP on Є–net points

  26. To Show:SDP value (=finite) > (1-ε)SDP value (=) SDP Vectors for =finite = Corresponding vectors in Є–net LP variables { μ …}? Problem : The inner products of vectors corresponding to a clause P might not be in the integral hull. ( For example : 3 arbitrary vectors in a Є–net are not guaranteed to satisfy triangle inequality ) The initial SDP solution satisfied all the constraints

  27. From STEP 2, We have discarded clauses for which inner products are not preserved within Є Discarding a clause P Forget about constraints corresponding to P • STEP 1 : Dimension Reduction • Project the SDP solution along • d =1/ Є4 random directions. STEP 2 : Throw away Discard clauses for which the corresponding inner products are not preserved within Є. =‘ = New instance • STEP 3 : Discretization • Pick an Є–netfor the • d dimensional sphere • Move every variable to the nearest point in the Є–net • =finite = discretized instance Discretization changes inner product by Є For every remaining clause, all inner products are within 2Єof what it was.

  28. Smoothing Operation Consider the inner products corresponding to a single clause P Canonical SDP Solution Uniform Distribution over all Integral solutions. Example: Va,0∙Va,0 = Va,1∙Va,1 = ½ Va,0∙Vb,0 = Va,0∙Vb,1 = Va,1∙Vb,0 = Va,1∙Vb,1 = 1/4 ЄX + Integral Hull Є–net Solution SDP Vectors for =finite = Corresponding vectors in Є–net (1-Є) X = Є Final SDP solution SDP Objective value remains roughly the same.

  29. Conclusions • Rounding stronger SDPs. • More efficient rounding? Can this SDP be solved in constant dimensional space directly? • Integrality gaps for stronger SDP relaxation of Unique Games

  30. Thank You

  31. Good finite Models from SDP solutions – Dimension Reduction & Discretization Idea: use almost SDP solution and do surgery Theorem: SDP value (=finite) > SDP value (=) ¦-CSP Instance =finite ¦-CSP Instance = identify variables with same vectors compute Dimension Reduction almost SDP solution for = almost SDP solution for = Discretize SDP solution for = Project on random low dimensional subspace Move vectors to closest point on ²-net finite number of different vectrs Rn Rd

  32. Constraint Satisfaction Problems (CSP) CSP ¦finite set of allowed types of constraints Á : [q]k {0,1} (alphabet [q], arity k)e.g. ¦ = { 3XOR, 3SAT, 3NAE} Examples: Max-Cut, Max-3SAT,… PCP Theorem: NP-hard to distinguish opt(=)=1 and opt(=)<0.9 (even for constant k and q) ¦-CSP Instance = • variables x1,…,xn • list of constraints Á of type ¦ on subsets of variables Approximation Algorithms: Goemans-Williamson, Zwick, CMM, … Goal: Find assignment x 2 [q]n so as to maximize fraction of satisfied constraints opt(=)

More Related