Structure identification in relational data

doi:10.1016/0004-3702(92)90009-M

Artificial Intelligence

Volume 58, Issues 1–3, December 1992, Pages 237-270

https://doi.org/10.1016/0004-3702(92)90009-M Get rights and content

Abstract

This paper presents several investigations into the prospects for identifying meaningful structures in empirical data, namely, structures permitting effective organization of the data to meet requirements of future queries. We propose a general framework whereby the notion of identifiability is given a precise formal definition similar to that of learnability. Using this framework, we then explore if a tractable procedure exists for deciding whether a given relation is decomposable into a constraint network or a CNF theory with desirable topology and, if the answer is positive, identifying the desired decomposition. Finally, we address the problem of expressing a given relation as a Horn theory and, if this is impossible, finding the best k-Horn approximation to the given relation. We show that both problems can be solved in time polynomial in the length of the data.

References (31)

R. Dechter
Decomposing a relation into a tree of binary relations
J. Comput. Syst. Sci.
(1990)
R. Dechter et al.
Network-based heuristics for constraint-satisfaction problems
Artif. Intell.
(1987)
R. Dechter et al.
Tree clustering for constraint networks
Artif. Intell.
(1989)
W.F. Dowling et al.
Linear time algorithms for testing the satisfiability of propositional Horn formula
J. Logic Program
(1984)
U. Montanari
Networks of constraints, fundamental properties and applications to picture processing
Inf. Sci.
(1974)
D. Angluin
Queries and concept learning
Mach. Learn.
(1988)
D. Angluin et al.
Learning conjunctions of Horn clauses
S. Arnborg
Efficient algorithms for combinatorial graphs with bounded decomposability—a survey
BIT
(1985)
C. Beeri et al.
On the desirability of acyclic database schemes
J. ACM
(1983)
A. Blumer et al.
Learnability and the Vapnik-Chervonenkis dimension
J. ACM
(1989)

R.K. Brayton et al.

Multilevel logic synthesis

C.K. Chow et al.

Approximating discrete probability distributions with dependence trees

IEEE Trans. Inf. Theor.

(1968)

S.A. Cook

The complexity of theorem-proving procedures

R. Dechter

Constraint networks

R. Dechter et al.

The complexity of finding all solutions

Cited by (108)

Learning Horn envelopes via queries from language models
2023, International Journal of Approximate Reasoning
We present an approach for systematically probing a trained neural network to extract a symbolic abstraction of it, represented as a Boolean formula. We formulate this task within Angluin's exact learning framework, where a learner attempts to extract information from an oracle (in our work, the neural network) by posing membership and equivalence queries. We adapt Angluin's algorithm for Horn formula to the case where the examples are labelled w.r.t. an arbitrary Boolean formula in CNF (rather than a Horn formula). In this setting, the goal is to learn the smallest representation of all the Horn clauses implied by a Boolean formula—called its Horn envelope—which in our case correspond to the rules obeyed by the network. Our algorithm terminates in exponential time in the worst case and in polynomial time if the target Boolean formula can be closely approximated by its envelope. We also show that extracting Horn envelopes in polynomial time is as hard as learning CNFs in polynomial time. To showcase the applicability of the approach, we perform experiments on BERT based language models and extract Horn envelopes that expose occupation-based gender biases.
Dualization in lattices given by ordered sets of irreducibles
2017, Theoretical Computer Science
Dualization of a monotone Boolean function on a finite lattice can be represented by transforming the set of its minimal 1 values to the set of its maximal 0 values. In this paper we consider finite lattices given by ordered sets of their meet and join irreducibles (i.e., as a concept lattice of a formal context). We show that in this case dualization is equivalent to the enumeration of so-called minimal hypotheses. In contrast to usual dualization setting, where a lattice is given by the ordered set of its elements, dualization in this case is shown to be impossible in output polynomial time unless P = NP. However, if the lattice is distributive, dualization is shown to be possible in subexponential time.
Disjunctive closures for knowledge compilation
2014, Artificial Intelligence
The knowledge compilation (KC) map [1] can be viewed as a multi-criteria evaluation of a number of target classes of representations for propositional KC. Using this map, the choice of a class for a given application can be made, considering both the space efficiency of it (i.e., its ability to represent information using little space), and its time efficiency, i.e., the queries and transformations which can be achieved in polynomial time, among those of interest for the application under consideration. When no class of propositional representations offers all the transformations one would expect, some of them can be left implicit. This is the key idea underlying the concept of closure introduced here: instead of performing computationally expensive transformations, one just remembers that they have to be done. In this paper, we investigate the disjunctive closure principles, i.e., disjunction, existential quantification, and their combinations. We provide several characterization results concerning the corresponding closures. We also extend the KC map with new propositional languages obtained as disjunctive closures of several incomplete propositional languages, including the well-known KROM (the CNF formulae containing only binary clauses), HORN (the CNF formulae containing only Horn clauses), and AFF (the affine language, which is the set of conjunctions of XOR-clauses). Each introduced language is evaluated along the lines of the KC map.
On minimal constraint networks
2012, Artificial Intelligence
In a minimal binary constraint network, every tuple of a constraint relation can be extended to a solution. The tractability or intractability of computing a solution to such a minimal network was a long standing open question. Dechter conjectured this computation problem to be NP-hard. We prove this conjecture. We also prove a conjecture by Dechter and Pearl stating that for $k ⩾ 2$ it is NP-hard to decide whether a single constraint can be decomposed into an equivalent k-ary constraint network. We show that this holds even in case of bi-valued constraints where $k ⩾ 3$ , which proves another conjecture of Dechter and Pearl. Finally, we establish the tractability frontier for this problem with respect to the domain cardinality and the parameter k.
On the Boolean connectivity problem for Horn relations
2010, Discrete Applied Mathematics
Gopalan et al. studied in [P. Gopalan, P.G. Kolaitis, E.N. Maneva, C.H. Papadimitriou, The connectivity of Boolean satisfiability: computational and structural dichotomies, in: Proceedings of the 33rd International Colloquium on Automata, Languages and Programming, ICALP 2006, 2006, pp. 346–357] and [P. Gopalan, P.G. Kolaitis, E.N. Maneva, C.H. Papadimitriou, The connectivity of Boolean satisfiability: computational and structural dichotomies, SIAM J. Comput. 38 (6) (2009) 2330–2355] connectivity properties of the solution-space of Boolean formulas, and investigated complexity issues on the connectivity problems in Schaefer’s framework. A set $S$ of logical relations is Schaefer if all relations in $S$ are either bijunctive, Horn, dual Horn, or affine. They first conjectured that the connectivity problem for Schaefer is in $P$ . We disprove their conjecture by showing that there exists a set $S$ of Horn relations such that the connectivity problem for $S$ is $co NP$ -complete. We also investigate a tractable aspect of Horn and dual Horn relations with respect to characteristic sets.
Information loss in knowledge compilation: A comparison of Boolean envelopes
2010, Artificial Intelligence
Since Selman and Kautz's seminal work on the use of Horn approximation to speed up the querying of knowledge bases, there has been great interest in Boolean approximation for AI applications. There are several Boolean classes with desirable computational properties similar to those of the Horn class. The class of affine Boolean functions, for example, has been proposed as an interesting alternative to Horn for knowledge compilation. To investigate the trade-offs between precision and efficiency in knowledge compilation, we compare, analytically and empirically, four well-known Boolean classes, and their combinations, for ability to preserve information. We note that traditional evaluation which explores unit-clause consequences of random hard 3-CNF formulas does not tell the full story, and we complement that evaluation with experiments based on a variety of assumptions about queries and the underlying knowledge base.

View all citing articles on Scopus

^☆: This work was supported in part by the Air Force Office of Scientific Research grant AFOSR 900136, NSF grant IRI-9157636, GE Corporate R&D and Micro grant 91-125.

View full text

Structure identification in relational data☆

Abstract

J. Comput. Syst. Sci.

Artif. Intell.

Artif. Intell.

J. Logic Program

Inf. Sci.

Queries and concept learning

Mach. Learn.

Learning conjunctions of Horn clauses

Efficient algorithms for combinatorial graphs with bounded decomposability—a survey

BIT

On the desirability of acyclic database schemes

J. ACM

Learnability and the Vapnik-Chervonenkis dimension