Enumeration of minimal transversals of hypergraphs of bounded VC-dimension
Abstract
We consider the problem of enumerating all minimal transversals (also called minimal hitting sets) of a hypergraph . An equivalent formulation of this problem known as the transversal hypergraph problem (or hypergraph dualization problem) is to decide, given two hypergraphs, whether one corresponds to the set of minimal transversals of the other. The existence of a polynomial time algorithm to solve this problem is a long standing open question. In [14], the authors present the first sub-exponential algorithm to solve the transversal hypergraph problem which runs in quasi-polynomial time, making it unlikely that the problem is (co)NP-complete.
In this paper, we show that when one of the two hypergraphs is of bounded VC-dimension, the transversal hypergraph problem can be solved in polynomial time, or equivalently that if is a hypergraph of bounded VC-dimension, then there exists an incremental polynomial time algorithm to enumerate its minimal transversals. This result generalizes most of the previously known polynomial cases in the literature since they almost all consider classes of hypergraphs of bounded VC-dimension. As a consequence, the hypergraph transversal problem is solvable in polynomial time for any class of hypergraphs closed under partial subhypergraphs. We also show that the proposed algorithm runs in quasi-polynomial time in general hypergraphs and runs in polynomial time if the conformality of the hypergraph is bounded, which is one of the few known polynomial cases where the VC-dimension is unbounded.
1 Introduction
A hypergraph is a couple where is a finite set called the vertices and is a family of subsets of called the hyperedges of the hypergraph. By abuse of notation, we often treat a hypergraph as its set of hyperedges when the set of vertices is clear from the context, and we say that a set of vertices belongs to if is a hyperedge of .
A set is called a transversal (or a hitting set) of if for all . A transversal is said to be minimal if it does not contains any other transversal.
The set of minimal transversals of forms another hypergraph on the same vertex set denoted by and referred to as the transversal hypergraph or the dual hypergraph of [1]. In this paper we are interested in the problem of finding given .
Trans-Enum
Input : A hypergraph .
Output : All minimal transversals of , i.e. .
Since could be exponentially larger than , this problem falls into the category of enumeration problems. To measure the complexity of algorithms that solving this kind of problems, we usually take into account both the input size (the size of ) and the output size (the size of ). With this paradigm (called the output-sensitive approach), an algorithm is said to be output-polynomial if its running time is a polynomial in and . We say that an algorithm runs in incremental polynomial time, if it can find minimal transversals in time polynomial in and . The problem admits an incremental polynomial time algorithm if and only if the following problem can be solved in polynomial time (in a classical sense):
Trans-Hyp
Input : Two hypergraphs and on the same vertex set with .
Output : Either answer that or find
We assume throughout the paper that is Sperner i.e. no hyperedge of contains another hyperedge. This assumption can be made without loss of generality since otherwise has the same minimal transversals as the restriction to its inclusion-wise minimal hyperedges. For Sperner hypergraphs, it is well known that and forms a duality relationship in the sense that if and only if [1]. Trans-Hyp problem corresponds to decide whether two hypergraphs are dual, and to find a counter-example otherwise. It has been shown in [2] that the simple decision version of this problem (without requiring a counter-example) is equivalent.
Trans-Hyp problem has been extensively studied, due to its equivalence to many other important problems (see e.g. [2, 7, 9, 10, 15, 16, 12, 19]). It is a long standing open question to decide whether Trans-Hyp can be solved in polynomial time. The best known algorithms to solve it runs in quasi-polynomial time where and [14]. The problem is then very unlikely to be NP-hard. The VC-dimension of a hypergraph was introduced in [20] and has been shown to be a important parameter for many different applications. Given a class of hypergraph , the Trans-Hyp problem restricted to consists in all instances of Trans-Hyp such that at least one of or belongs to . The main result of this paper is that Trans-Hyp can be solved in polynomial time in hypergraphs of bounded VC-dimension, or equivalently, that Trans-Enum can be solved in incremental polynomial time for hypergraphs of bounded VC-dimension.
Theorem 0.
Let be a hypergraph with and let . One can decide in time whether and find otherwise.
Many different polynomial time algorithms for Trans-Hyp have been designed for special classes of hypergraphs in the literature. While more efficient algorithms are generally presented, Theorem 1 generalizes many of those results since we observe that considered classes often have bounded VC-dimension. Among the already known polynomial cases directly covered by Theorem 1 we can cite:
- β’
- β’
- β’
We observe that all those classes are closed under partial subhypergraphs. A partial subhypergraph of a hypergraph is a hypergraph obtained from by selecting a subset of hyperedges , a subset of vertices and considering the hyperedges of restricted to , i.e. the hypergraph . As an important corollary of Theorem 1 we obtain the following
Corollary 1.
Trans-Hyp can be solved in polynomial time in any proper class of hypergraphs closed under partial subhypergraph.
One of the most general class for which we already know that Trans-Hyp is solvable in polynomial time and which is not covered by Theorem 1 is the class of -conformal hypergraphs [16]. While this class is not closed under partial subhypergraph and does not have a bounded VC-dimension we show that the algorithm developed in this paper runs also in polynomial time if the hypergraph is -conformal. Up to our knowledge the two main cases for which the polynomiality cannot be directly deduced from the results present in this paper are the class of -degenerate hypergraphs [8] and the class of -exact hypergraphs [11].
2 Preliminaries
A trace on is a couple with , and . The size of trace is defined as and it is called a -trace if . We say that a subset realizes a trace if and we denote by the set of -traces realized by on . When the set is clear from the context, we will simply use instead of For a hypergraph we denote by the set of traces realized by its hyperedges. In other words, a k-trace belongs to if there exists a hyperedge such that . A subset is k-compatible with if i.e. for each k-subset there exists such that . Given a hypergraph and , the -extension of , is the hypergraph on whose hyperedges are all subsets of -compatible with , i.e. . Since in particular every hyperedge of is -compatible with , .
A subset of vertices is scattered in if for all , there exists such that i.e. if for all . The VC-dimension of , , is the size of its largest scattered set. Using the trace definition, if for all -subsets of there exists such that . One of the most important property of hypergraphs of bounded dimension is given by the SauerβShelah Lemma.
Lemma 2.
SauerβShelah Lemma If is of VC-dimension strictly less than , then
Corollary 3.
If is of VC-dimension strictly less than then
Proof.
Since by definition, has exactly the same -traces than a subset of vertices size is scattered in if and only if it is scattered in . Since scatters no subset of size , . The result follows from Lemma 2. β
Given a hypergraph and a subset of hyperedges , the hypergraph is called the partial hypergraph of induced by . Given a subset of vertices the hypergraph is the subhypergraph induced by . Given and the partial subhypergraph induced by and is the hypergraph . A partial subhypergraph of that does not contain isolated vertices (i.e. vertices that do not belong to any hyperedge) is a partial hypergraph of a subhypergraph of or equivalently a subhypergraph of a partial hypergraph of .
3 Main algorithm
In this section we assume that we are given a hypergraph with and a hypergraph . We want to decide whether or equivalently whether and find a new minimal transversal in otherwise. We present Algorithm 1 to solve this problem whose running time is as stated by Theorem 1.
Notice that we can assume that since otherwise the answer is no, and we can easily find a new minimal transversal . Indeed assume that there exists such that is not a minimal transversal of , i.e. there exists such that is also a transversal of , then any minimal transversal of included in belongs to .
The algorithm will actually either answer that or find a new minimal transversal otherwise. In the latter case, given , one can easily find a new minimal transversal of . Indeed it is enough to output any minimal transversal of included in .
We are trying to find such that . The main strategy is based on the observation that if such a exists, then either is -compatible with (i.e. ) or it realizes a -trace that is not in . It is easy to check whether there exists a which satisfy the later case in general hypergraphs, and we show that the former case can be checked in polynomial time when . The strategy is summarized in Algorithm 1.
The first loop of Algorithm 1 try to find a minimal transversal i.e. a minimal transversal of which realizes a -trace . To do so we simply go over all -traces that are not in and check for each one whether it is realizable by a minimal transversal of . There are at most such -traces and we will show in Corollary 7 that we can check in polynomial time whether a -trace is realizable by a minimal transversal of .
The second loop of Algorithm 1 try to find a minimal transversal of which is -compatible with and that does not belong to . For this, we first generate from and we try every possible hyperedge of that is not included in a hyperedge of . In general the size of may be exponentially larger than the size of , but we prove in Proposition 10 that when , its size is polynomial and we can compute it in polynomial time.
To check the condition on line 3, we need to test whether has a minimal transversal containing and excluding which can be done in polynomial time when is of bounded size as we will see in section 3.1. The remaining part is to compute . We will see in section 3.2 that this can be done in polynomial time whenever is bounded.
Theorem 4.
For any , Algorithm 1 correctly returns "Yes" if and a set otherwise.
Proof.
Let us show first that if a set is returned by the algorithm, then .
Assume first that is returned in line 4, i.e. that the condition in line 3 is true. The condition imposes that and since realizes a -trace that does not belong to , cannot belong to since otherwise all -traces realized by would belong to . So . Suppose now that is returned in line 7. Notice that we impose in line 5 that is contained in no hyperedge of . In particular, cannot belong to . Together with the condition in line 6, is returned in line 7 only if . So if then the algorithm answer "Yes" since otherwise it would return a set .
Assume now that the algorithm answer "Yes" and let us show that . Assume for contradiction that there exists . If then and then, would be returned in the loop in line 5. Otherwise there is a -trace . But then, would be a minimal transversal of which realizes a trace and such a minimal transversal of would be returned in line 4. Notice that it wouldnβt be necessarily that would be returned, but any that realizes . β
3.1 Finding a minimal transversal satisfying a -trace
In this section, we show that the condition in line 3 of Algorithm 1 can be check in polynomial time. More precisely, given a -trace we show that we can check in time whether there exists a minimal transversal of realizing . When is a constant, is of constant size and the above mentioned complexity is polynomial. To do so, we reduce the problem to deciding whether is a sub-transversal in a subhypergraph of . A set of vertices is a sub-transversal of if there exists a minimal transversal such that . In [5], the authors proved that one can check in polynomial time whether a subset of vertices of bounded size is a sub-transversal.
Lemma 5.
[5] Given a subset of vertices, one can check in time whether is a sub-transversal of .
Lemma 6.
Let be a -trace on and let . Then, there exists a minimal transversal of realizing if and only if is sub-transversal of the subhypergraph
Proof.
Observe first that if and only if is a minimal transversal of such that . Assume first that is a sub-transversal of . There exists such that . Since is a minimal transversal of it is a also a minimal transversal of . Now since and we have and so realizes . Assume now that there exists such that realizes . We have and so is included in . Therefore, is a minimal transversal of containing . β
Corollary 7.
Given a -trace on can check in time whether there exists a minimal transversal of realizing .
3.2 Computing
In this section we show that when , can be computed in time
For , let us denote by the set and let be the hypergraph . We iteratively build from . For each we check whether and violate a -trace of -trace .
Lemma 8.
Proof.
As notices previously for any hypergraph , so . So let us show that . Let . Since the only -traces realized by on is the -trace . Now since , there exists a hyperedge which realizes the -trace , i.e such that . But since both and , we have , i.e. is a hyperedge of . β
Lemma 9.
Let and let , then .
Proof.
Let . Let us show that . Let with . Since there exists such that . Let . Since , . Since is a hyperedge of , is a hyperedge of and so the trace .
β
Proposition 10.
If then can be computed in time
Proof.
To compute we iteratively compute from starting by up to . By Lemma 8 we start from . Then for each we compute from in the following way :
For each Do:
Add to if
Add to if .
By Lemma 9 each hyperedge of is found by the above procedure. Now observe that for any , the hypergraph is a subhypergraph of and so . So, by Corollary 3, . Now, for a given we need to check whether (resp. ) belong to i.e whether (resp. ). We claim that this can be checked in time . Let be such that . Since and since , there exists such that . Now there exists such that and since , we have . So the -trace (resp. ). So if (resp. ) realizes a trace (resp. ), must contain . Therefore, to check whether (resp. ) we only have to check for each of the subsets of of size that contain , whether .
Hence, is computed in since we check for each of the hyperedge of whether and belong to which can be done in . So in total can be computed in time .
β
We are now ready to prove Theorem 1
See 1
Proof.
From Theorem 4 Algorithm 1 returns "Yes" if and only if and otherwise. By the duality property, the algorithm returns "Yes" if and only if . Assume now that the algorithm returns an element . Since we assumed that all hyperedges of are transversals of , does not contain any heyperedge of and thus is a transversal of . Observe furthermore that contains no hyperedge of since otherwise would not be a transversal of . So to output a minimal transversal it is enough to compute any minimal transversal of contained in which can be done in . Let us show now that Algorithm 1 runs with the announced complexity. The for loop in line 2 goes over all -traces . There is at most such traces and for each -trace the condition in line 3 can be checked in time by Corollary 7. So the first for loop take in total .
For the second loop, by Lemma 10, we can compute in time . Since for each one can check whether in time the total time taken by the for loop in line 2 is .
β
3.3 Consequences
The first immediate consequence of Theorem 1 is obtained when the VC-dimension is bounded, i.e. when is a constant.
Corollary 11.
Trans-Hyp problem is solvable in polynomial time in hypergraph classes of bounded VC-dimension.
Even if several quasi-polynomial time algorithms are already known for general hypergraphs, we observe Algorithm 1 runs also in quasi-polynomial time.
Corollary 12.
Algorithm 1 runs in time .
Proof.
Notice that if a set of size is scattered by , then contains at least hyperedges, and so we have . So by Theorem 1, we obtain the result. β
Another important property of Algorithm 1 is that it runs in polynomial time on hypergraphs of bounded conformality. The conformality of hypergraph was introduced in [1], a hypergraph is said to be -conformal if any minimal subset of vertices that is not included in any hyperedge of is of size at most . In [18], the authors prove that Trans-Hyp is solvable in polynomial time in -conformal hypergraphs. Although -conformal hypergraphs can have arbitrarly large VC-dimension, Algorithm 1 runs in polynomial time for hypergraphs of bounded conformality. In the proof of Theorem 1 the VC-dimension is used to bound the size and to be able to compute it in polynomial time. The following proposition assert that if is -conformal, then any hyperedge of is a subset of an hyperedge of . Hence, even though the size of may be exponential in , we donβt need to compute it since the for loop in line 5 of Algorithm 1 only runs through hyperedges of not included in any hyperedge of .
Proposition 13.
Let be a -conformal hypergraph. If , there exists such that .
Proof.
Assume that there exists such that is not included in any hyperedge of . Let be a minimal subset that is contained in no hyperedge of . is well defined since itself is contained in no hyperedge of . By definition of the conformality, we have . Let be any superset of of size . Then the -trace is realized by but does not belong to since while no hyperedge of contains . So realizes a -trace which does not belongs to which is in contradiction with .
β
Corollary 14.
Agorithm 1 runs in time on -conformal hypergraphs
4 Hypergraph classes closed under subhypergraphs
As already mentioned in the introduction, a consequence of Theorem 1 or more precisely of corollary 11 is that Trans-Hyp is solvable in polynomial time in any proper class of hypergraph closed under partial subhypergraph. Indeed, any class of hypergraphs that forbids a hypergraph as partial subhypergraph is of bounded dimension and thus, Algorithm 1 can be used to solve the Trans-Hyp problem. We also show in this section that such a result is hopeless for the other notions of subhypergraph except if one can solve the Trans-Hyp problem in general hypergraphs in polynomial time.
4.1 Classes closed under partial subhypergraphs.
Lemma 15.
If , then contains all hypergraphs on at most vertices as partial subhypergraph.
Proof.
Let be a hypergraph with . Since , has a scattered set of vertices of size . Now since is scattered, for every there exists such that and the partial subhypergraph of is a copy of . β
Proposition 16.
Any proper class of hypergraphs which is closed under partial subhypergraphs has a bounded VC-dimension.
Proof.
Let a hypergraph with the minimum number of vertices which is not in . Such a hypergraph exists since is a proper class of hypergraphs (i.e. it does not contain all hypergraphs). Let us show that the VC-dimension of is bounded by . Indeed let . Since is closed under partial subhypergraph and since , does not contain as partial subhypergraph. Hence, by Lemma . β
See 1
4.2 Other subhypergraphs types
Several types of subhypergraphs have been defined in the literature namely, partial hypergraphs, subhypergraphs, partial subhypergraphs, edge-induced subhypergraphs and restrictions. Even if the the concepts of partial hypergraphs, subhypergraphs and partial subhypergraphs have been already defined, we recall them for comparison to the other notions. For a hypergraph ,
-
β’
a partial hypergraph of is a hypergraph , for ;
-
β’
a subhypergraphs of is a hypergraph , for ;
-
β’
a partial subhypergraphs of is a hypergraph , for and ;
-
β’
an edge-induced subhypergraphs of is a hypergraph , for ;
-
β’
a restriction of is a hypergraph , for .
We show that a result similar to the one obtained in Corollary 1 for the other notions of subhypergraphs would imply a polynomial time algorithm for the general problem. More precisely we focus on the class of hypergraphs that forbids a specific hypergraph as subhypergraph. Given a hypergraph , we denote by , , , the set of hypergraphs that forbid respectively as subgraph, edge-induced subgraph and restriction. For the case of edge-induced subhypergraphs, for any hypergraph , we prove that the Trans-Hyp problem in the class is as hard as in the general case.
For the case of restrictions and induced subhypergraphs, the difficulty of Trans-Hyp in the classes and depends of the hypergraph . For exemple if is the complete hyperegraph on vertices (the hypergraph that contains all subsets as hyperedges), then the class is precisely the class of hypergraph of VC-dimension strictly smaller than and Theorem 1 the problem is polynomial. However, if the hypergraph is sparse enough, we prove that the Trans-Hyp problem restricted to the class is as hard as in general hypergraphs.
For the case of restrictions, if is the empty hypergraph on vertices (the hypergraph without any hyperedge), then the class is the class of hypergraph for which the maximum independent set is strictly smaller than . Since maximal independent sets are the complements of minimal transversals, one can enumerate all subsets of size smaller than and check whether the complement is a minimal transversal. Therefore, if is the empty hypergraph, then Trans-Hyp problem in is polynomial. In any other case, we prove that Trans-Hyp problem restricted to is as hard as the general problem.
In order to prove the results of this section, we will use two different reductions. Given a hypergraph and an integer , we define to be the hypergraph obtained from by adding new vertices to every hyperedges, and to be the hypergraph obtained from by adding new hyperedges of size corresponding to all -subsets of vertices of plus a new vertex .
-
β’
-
β’
We now show how the enumeration of minimal transversal of can be reduced to the enumeration of minimal transversal of and .
Proposition 17.
Let be an integer and be a hypergraphs, then .
Proof.
Let . Then either and is also a minimal transversal of or contains at least one vertex . Since belongs to every hyperedges of is a transversal and my minimality of we have . So . Now since every minimal transversal of is also a minimal transversal of and any belongs to every hyperedge of , we have . β
Proposition 18.
Let be an integer, be a hypergraphs, then where only contains subsets of size larger than .
Proof.
Let . Assume first that the only vertex belongs to . Then must minimally cover the hyperedges that donβt contain i.e. must be a minimal transversal of since the hyperedges that contain are exactly . Assume now that and assume that . Let be a subset of vertices included in the complement of such that , then is a hyperedge of whose intersection with would be empty contradicting the fact that is a transversal of . So if then either and is a minimal transversal of or . Assume now that ) with and let us show that . Since is a minimal transversal of and since belongs to every hyperedge of , is a transversal of . Let us show that it is minimal. Observe first that since , cannot be a transversal since its complement contains at least one -subset of and the hyperedge would be avoided by . Now if there exists such that , since does not belong to any original hyperedge of , would be a transversal of contradicting the minimality of . β
Lemma 19.
Let be a fixed hypergraph on vertices, and let be a hypergraph on at least vertices, then :
-
1.
is edge-induced -free.
-
2.
If has at least one hyperedge, is restriction -free.
-
3.
If there exists such that has strictly less than hyperedges of size , then is subhypergrah -free.
Proof.
1) Since every hyperedges of contains the set of additional vertices, any edge-induced subgraph of contains at least these vertices. Since has only vertices, no edge-induced subgraph of can be isomorphic to .
2) Let . Since is included in every hyperedges of , if , the restriction of to contains no hyperedge. So any non empty restriction of must contain and so contains at least vertices. Since has exactly vertices, is restriction -free.
3) Assume that there exists such that has strictly less than hyperedges of size and let be a subset of size . Let be the extra vertex added in not present in . If does not contain , any subset of of size will forms a hyperedge with . Since does not belong to , is an hyperedge of the subhypergraph induced by . So the subhypergraph induced by contains hyperedges of size and cannot be isomorphic to . Assume now that . Since we assumed that has at least vertices, has at least vertices and there exists a vertex . Now for any subset of , of size , there exists a hyperedge in and so the subhypergraph induced by has as hyperedge. So the subhypergraph induced by contains at least hyperedges of size and cannot be isomorphic to . β
Theorem 20.
Let be a hypergraph, then the restriction of the Trans-Hyp problem to the following classes of hypergraph is as hard as the Trans-Hyp problem in general hypergraphs:
-
β’
-
β’
if is different from the empty hypergraph.
-
β’
if there exists such that has less than hyperedges of size .
Proof.
By lemma 19 it is sufficient to prove that for a given integer , one can reduce in polynomial time the enumeration of minimal transversal of to the enumeration of minimal transversal of and .
Assume that one can enumerate the minimal transversal of in . By lemma 17, . So one can simply check for each whether is a minimal transversal of . Since and , the total running time is .
Assume now that one can enumerate the minimal transversal of in . By lemma 18, where only contains subsets of size larger than . One can start by enumerating all minimal transversal of of size larger than in a brute-force way by checking all the such subsets. Then for every with that contains , output (where is the additional vertex of ). Since and , the total running time is .
β
References
- [1] Claude Berge. Hypergraphs: combinatorics of finite sets. Number v. 45 in North-Holland mathematical library. North Holland : Distributors for the U.S.A. and Canada, Elsevier Science Pub. Co, Amsterdam ; New York, 1989.
- [2] J.C. Bioch and T.Β Ibaraki. Complexity of identification and dualization of positive boolean functions. Information and Computation, 123(1):50β63, November 1995.
- [3] E.Β Boros, V.Β Gurvich, K.Β Elbassioni, and L.Β Khachiyan. An efficient incremental algorithm for generating all maximal independent sets in hypergraphs of bounded dimension. Parallel Processing Letters, 10(04):253β266, December 2000.
- [4] Endre Boros, Khaled Elbassioni, Vladimir Gurvich, and Leonid Khachiyan. Generating Maximal Independent Sets for Hypergraphs with Bounded Edge-Intersections. In Gerhard Goos, Juris Hartmanis, Jan van Leeuwen, and MartΓn Farach-Colton, editors, LATIN 2004: Theoretical Informatics, volume 2976, pages 488β498. Springer Berlin Heidelberg, Berlin, Heidelberg, 2004.
- [5] Endre Boros, Vladimir Gurvich, and PeterΒ L. Hammer. Dual subimplicants of positive boolean functions. Optimization Methods and Software, 10(2):147β156, 1998.
- [6] Thomas Eiter and Georg Gottlob. Identifying the minimal transversals of a hypergraph and related problems. SIAM Journal on Computing, 24(6):1278β1304, December 1995.
- [7] Thomas Eiter and Georg Gottlob. Hypergraph Transversal Computation and Related Problems in Logic and AI. In G.Β Goos, J.Β Hartmanis, J.Β van Leeuwen, Sergio Flesca, Sergio Greco, Giovambattista Ianni, and Nicola Leone, editors, Logics in Artificial Intelligence, volume 2424, pages 549β564. Springer Berlin Heidelberg, Berlin, Heidelberg, 2002.
- [8] Thomas Eiter, Georg Gottlob, and Kazuhisa Makino. New results on monotone dualization and generating hypergraph transversals. Proceedings of the thiry-fourth annual ACM symposium on Theory of computing, pageΒ 9, 2002.
- [9] Thomas Eiter, Kazuhisa Makino, and Georg Gottlob. Computational aspects of monotone dualization: A brief survey. Discrete Applied Mathematics, 156(11):2035β2049, June 2008.
- [10] Khaled Elbassioni, Kazuhisa Makino, and Imran Rauf. Output-Sensitive Algorithms for Enumerating Minimal Transversals for Some Geometric Hypergraphs. In Amos Fiat and Peter Sanders, editors, Algorithms - ESA 2009, volume 5757, pages 143β154. Springer Berlin Heidelberg, Berlin, Heidelberg, 2009.
- [11] Khaled Elbassioni and Imran Rauf. Polynomial-time dualization of -exact hypergraphs with applications in geometry. Discrete Mathematics, 310(17-18):2356β2363, September 2010.
- [12] Khaled Elbassioni, Imran Rauf, and Saurabh Ray. A global parallel algorithm for enumerating minimal transversals of geometric hypergraphs. Theoretical Computer Science, 767:26β33, 2019.
- [13] Khaled Elbassioni, Imran Rauf, and Saurabh Ray. A global parallel algorithm for enumerating minimal transversals of geometric hypergraphs. Theoretical Computer Science, 767:26β33, May 2019.
- [14] MichaelΒ L. Fredman and Leonid Khachiyan. On the complexity of dualization of monotone disjunctive normal forms. Journal of Algorithms, 21(3):618β628, November 1996.
- [15] Dimitrios Gunopulos, Heikki Mannila, Roni Khardon, and Hannu Toivonen. Data mining, hypergraph transversals, and machine learning (extended abstract). In Proceedings of the sixteenth ACM SIGACT-SIGMOD-SIGART symposium on Principles of database systems - PODS β97, pages 209β216, Tucson, Arizona, United States, 1997. ACM Press.
- [16] Leonid Khachiyan, Endre Boros, Khaled Elbassioni, and Vladimir Gurvich. On the dualization of hypergraphs with bounded edge-intersections and other related classes of hypergraphs. Theoretical Computer Science, 382(2):139β150, August 2007.
- [17] Leonid Khachiyan, Endre Boros, Vladimir Gurvich, and Khaled Elbassioni. Computing many maximal independent sets for hypergraphs in parallel. Parallel Processing Letters, 17(02):141β152, June 2007.
- [18] Nina Mishra and Leonard Pitt. Generating all maximal independent sets of bounded-degree hypergraphs. In Proceedings of the tenth annual conference on Computational learning theory, COLT β97, pages 211β217, New York, NY, USA, July 1997. Association for Computing Machinery.
- [19] Ken Takata. A worst-case analysis of the sequential method to list the minimal hitting sets of a hypergraph. SIAM Journal on Discrete Mathematics, 21(4):936β946, January 2008.
- [20] V.Β N. Vapnik and A.Β Ya. Chervonenkis. On the uniform convergence of relative frequencies of events to their probabilities. Theory of Probability & Its Applications, 16(2):264β280, January 1971. Publisher: Society for Industrial and Applied Mathematics.