1 / 25

Approximability of Multiway Partition

Approximability of Multiway Partition. Yi Wu IBM Almaden Research Joint work with Alina Ene and Jan Vondrak. Definition of Problems. Graph Multiway Cut. Input: a graph with terminals. Graph Multiway Cut. Input: a graph with terminals.

hila
Download Presentation

Approximability of Multiway Partition

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. Approximability ofMultiway Partition Yi Wu IBM Almaden Research Joint work with AlinaEne and Jan Vondrak

  2. Definition of Problems

  3. Graph Multiway Cut Input: a graph with terminals.

  4. Graph Multiway Cut Input: a graph with terminals. Goal: remove minimum number of edges to disconnect the terminals.

  5. Graph Multiway Cut Input: a graph with terminals. Equivalent Goal: divide the graph into parts to minimize number of cross edges.

  6. Constraint Satisfaction Problem (CSP) with “” constraint 2 1 3 Equivalent Problem: assign to minimize the satisfied inequality.

  7. Approximability of Graph Multiway Cut • Upper bound • -approximation by [Calinescu-Karloff-Rabani,1998] • -approximation by [Karger-Klein-Stein-Thorup-Young, 1999] • Lower bound: assuming Unique Games Conjecture, • NP-hard to get better than -approximation. • an earth mover Linear Programming is optimal polynomial time approximation (the ratio is unknown). [Manokaran-Naor-Raghavendra-Schwartz,2008]

  8. Variant: Node Weighted Multiway Cut Goal: remove minimum number (weights) of nodes to disconnect the terminals.

  9. Variant:Hypergraph Multiway Cut (HMC) • Given a hypergraph and -terminals . Remove the minimum number of edges to disconnect • Approximation equivalent to Node Weighted Multiway Cut [Zhao-Nagamochi-Ibaraki 2005]. • Min-CSP with NAE (Not All Equal) constraint on the edges.

  10. Generalization:Submodular Multiway Partition • Given a ground set and some submodular function and terminals . Find set • = V Goal: minimize Hypergraph Multiway Cut is a special case[Zhao-Nagamochi-Ibaraki 2005]. A function is submodular if

  11. Another interesting SMP: Hypergraph Multiway Partition • Given a hypergraph graphwith -terminals. • partition the graph into parts. • The cost on each edge is the number of different parts it falls in.

  12. Relationship Submodular Multiway Partition Hypergraph Multiway cut = Node Weighted Multiway Cut Graph Multiway Cut Graph Multiway Cut Hypergraph Multiway Partition.

  13. Our Results

  14. Our Results (1) 4/3-approximation for 3-way submodular partion. • There is a -approximation for the general submodular multiway partition. • Previous Work: • -approximation by [Chekuri-Ene, 2011] • -approximation for node weighted/hypergraph multiway cut [Garg-Vazirani-Yannakakais,1994] Based on the half integrality of an LP.

  15. Overview of the algorithm • Lovasz Relaxation of Submodular Function: • Variables ( dimensional probability simplex) for any • Lovasz Relaxation is the expected value of for the following construction of : choosing random , assign to if • The rounding is not necessarily feasible as can be assigned to multiple • We can efficiently minimize ]

  16. The rounding algorithm • Choose a random • For every , set for (i.e, assign to the -th terminal. • Randomly set all the undecided terminals to a partition. To improve from to , the main technicality is analyzing step 3.

  17. Our result (2)matching UG-hardness • It is Unique Games-hard to get better than -approximation for hypergraph multiway cut. • Previous work: UG-hardness (from vertex cover). • We prove that assuming UGC, • The integrality gap of a basic LP (generalizing the earth mover LP) is the approximation threshold. • The integrality gap is . • The LP is also optimal for any CSPs contains constraint.

  18. The LP for Hypergraph Multiway Cut For hypergraph with terminals : • Variables: for every and , for every and . • Goal: • Constraint: • For every • For every and • For the -th terminal

  19. The LP for general Min-CSP For hypergraph and cost function on each • Variables: for every and , for every and . • Goal: • Constraint: • For every • For every and Optimal LP if the constraint contains .

  20. Our Results (3): matching oracle hardness • For the submodular multiway partition problem, it requires exponential number of queries on to get better than approximation. • We prove this by constructing symmetric gap of hypergraph multiway cut. Q: is it a coincident that the oracle hardness is the same as the Unique Games hardness?

  21. Symmetric gap for Hypergraph Multiway Cut • The graph is symmetric for any permutation from a permutation group . • Vertices are equivalent if there exists some permutation that • A (fractional) solution is symmetric if it is the same on all equivalent vertices. • Symmetric gap: Let I be a instance . Optimum solution (by independent rounding). Optimum Symmetric solution (by independent rounding).

  22. Why study symmetric gap? • Symmetric gap of a CSP implies an oracle hardness result for the submodular generalization of that CSP. • Symmetric gap for Max Cut -hardness for non-montone Submodular Function [Feige-Mirrokni-Vondrak, 2007] • Symmetric gap for NAZ (not all zero)  -hardness for Monotone Submodular Function with cardinality constraint [previously known Nemhauser-Wolsey, 1978] • Symmetric gap for Hypergraph Multiway Cut -Submodular Multiway Partition

  23. Our Results (4) • Q: is it a coincident that the oracle hardness is the same as the Unique Games hardness? • A: No. We prove that for any CSP instance, symmetric gap = LP integrality gap.

  24. Conclusion • We have a -approximation for the general submodular multiway partition problem. • oracle hardness and UG-hardness for the hypergraph multiway cut/Node Multiway Cut problem • Equivalence between LP gap and approximation threshold as well as oracle hardness for general CSPs.

  25. Open problem • The integrality gap of hypergraph multiway partition? • Between and • It is corresponding to an oracle hardness result for Symmetric submodular multiway partition.

More Related