R.M. Cameron-Jones and J.R. Quinlan, Avoiding Pitfalls When Learning Recursive Theories, Proceedings Thirteenth International Joint Conference on Artificial Intelligence 1993,1050-1055, Chambery, France.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....Foil Foil by Ross Quinlan and colleagues at the University of Sydney [ Qui90 ] is an efficient top down ILP learning algorithm based on heuristic search control. Foil was the first system to be interfaced to Mobal according to the external tool concept; currently, version 6 of the system [ CJQ93 ] can be used. Foil s special treatment of negative examples and the closed world assumption is taken care of in the interface module. mFoil mFoil by Saso Dzeroski from the Ljubljana AI Labs [ DB92 ] is a variant of Foil with a stochastic search control criterion; it is integrated in a fashion ....

R.M. Cameron-Jones and R. Quinlan. Avoiding pitfalls when learning recursive theories. In Proc. 13th International Joint Conference on Artificial Intelligence, pages 1050 -- 1055, San Mateo, CA, 1993. Morgan Kaufman.

....leads to the problem of determinate literals for empirical learning systems which rely on an entropy measure, as the system Foil [10] literals associated to functions do not allow to discriminate positive and negative examples and therefore, they are in general not introduced in the clause. In [3], the authors show that the program defining the predicate quicksort(l 1 ; l 2 ) cannot be learned, unless determinate literals are added to the body of the clause. In this paper, we propose a new approach for learning logic programs containing function symbols in a Constraint Logic Programming ....

R. M. Cameron-Jones, J.R. Quinlan, 1993. Avoiding Pitfalls When Learning Recursive Theories. Proceedings of IJCAI 93, Chamb'ery, France, Vol. 2, pp. 1050-1055.

....and it is based on an acceptability rate; both enable to reduce the influence of the order the predicates are learned. 1 Introduction Learning a logical definition of a single predicate from a set of positive and negative examples is a classical problem in Inductive Logic Programming (ILP) [14, 3, 9] referred to as single predicate learning (spl) In this paper, we consider the more general problem, called multiple predicate learning (mpl) that consists in finding a definition of a set of predicates from positive and negative examples. We show why putting together the definitions of each ....

....are learned: they have to prevent the creation of trivial clauses such as p(X) p(X) which cover all the positive examples and reject all the negative ones. This problem can be solved by learning only theories that are not recursive or by using biases using a well founded ordering of the domain [14, 3, 1]: for example, in the case of an unary predicate q, the restrictions induced by such an ordering can be stated as follows: if q(X) l 1 ; l n is the clause under construction, the literal q(Y ) is added to the body of the clause only if there exists l i = p(Z 1 ; Z k ) with p 2 base, ....

R. M. Cameron-Jones, J.R. Quinlan, 1993. Avoiding Pitfalls When Learning Recursive Theories. Proceedings of the Thirteen International Joint Conference on Artificial Intelligence, Chamb'ery, France, August 28 - September 3, 1993, Vol. 2, pp. 1050-1055.

....to produce complete programs, it induces strong restrictions. For example, the classical definition of transitive closure r of the relation r : f r (X,Y) r(X,Y) r (X,Y) r(X,Z) r (Z,Y) g cannot be learned if the relation r contains cycles. The improvement of this bias proposed in [8] still remains a strong restriction. Top down evaluation of the semantics SLD resolution is used by Clint for the validation of the learned program and for the validation of different steps of the construction. In order to ensure that a positive example is proved by the program, it tries to find ....

R. M. Cameron-Jones and J. R. Quinlan. Avoiding pitfalls when learning recursive theories. In Proceedings of the 13th International Joint Conference on Artificial Intelligence, pages 1050-- 1055. Morgan Kaufmann, 1993.

....all the positive examples into the background knowledge. This allows the algorithm to determine coverage by using these facts to unify with the recursive literals in the body of the clause. However, this introduces many problems, different systems handle them in a variety of ways. For example, FOIL[6] induces a partial ordering on the constants, and then requires at least one term in the recursive literal to descend or ascend this ordering. This ensures that there are no loops in the recursive call. CHILLIN[27] on the other hand, requires that at least one term in the recursive literal be a ....

M. Cameron-Jones and J. Quinlan. Avoiding pitfalls when learning recursive theories. In Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence. Morgan Kauffmann, San Mateo, CA, 1993.

.... include determinate literals (Muggleton Feng, 1990; Quinlan, 1991) We then tested the system on two ILP benchmarks: the finite element mesh design problem introduced by Dolsak and Muggleton (1992) and a selection of the list processing programs from Bratko (1990) previously used by Quinlan and Cameron Jones (1993). We compare Foidl s performance to Foil and to FFoil (Quinlan, 1996) a recent specialized version of Foil for learning single output functions without explicit negative examples. First order decision lists enable Foidl to achieve accuracy on the finite element mesh design problem that is ....

....use a logic representation, uses ILP techniques to learn patterns for information extraction (Califf Mooney, 1997) 1 2. 2 FOIL Since Foidl is based on Foil, we present a brief review of this important ILP system; see articles on Foil for a more complete description (Quinlan, 1990; Quinlan Cameron Jones, 1993; Cameron Jones Quinlan, 1994) 2 Foil learns a function free, first order, Hornclause definition of a target predicate in terms of itself and other background predicates. The input consists of extensional definitions of these predicates as a complete set of tuples of constants of specified ....

[Article contains additional citation context not shown here]

Cameron-Jones, R. M., & Quinlan, J. R. (1993). Avoiding pitfalls when learning recursive theories. In Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence, pp. 1050--1055 Chambery, France.

....they suffer from some problems. These problems are of two types: ffl the extensionality problem, which is specific to GOLEM and FOIL, is due to the incorrect definition of coverage, which is in turn due to working with an extensional 5. LEARNING MULTIPLE PREDICATES 239 background model (see [14, 2, 5]) ffl the order dependency problem when learning multiple predicates and or recursive predicate definitions. The first problem is specific to FOIL and GOLEM, whereas the second problem holds for the normal ILP setting in general; it is discussed in the context of multiple predicate learning in ....

....and GOLEM, whereas the second problem holds for the normal ILP setting in general; it is discussed in the context of multiple predicate learning in Section 5. 4. 2 The extensionality problem The problems with extensional models are well known and have already been published in the ILP literature [2, 5, 14]. In summary, to test whether a hypothesis H covers an example e, extensional methods use the following extensional coverage test. Definition 4.3 A hypothesis H extensionally covers an example e iff there exists a clause c 2 H and a substitution such that c is ground, head(c) e and ....

[Article contains additional citation context not shown here]

R.M. Cameron-Jones and J.R. Quinlan. Avoiding pitfalls when learning recursive theories. In Proceedings of the 13th International Joint Conference on Artificial Intelligence, pages 1050-- 8. CONCLUSIONS 253 1055. Morgan Kaufmann, 1993.

....of cyclic attributes useless for effective prediction. For instance an attribute the number of finite elements of my same neighbor s same neighbor would not make sense for prediction, as it references the node in question itself. Pitfalls of recursion in ILP are dealt with at length in [Cameron Jones Quinlan 93] A constructive induction system equipped with such operators might offer an alternative perspective 5 on ILP, possibly providing a more natural fit for data in object oriented representations or databases. 5 Conclusions, Related Work, and Further Research Incorporating the MDL Principle ....

Cameron-Jones R.M., Quinlan J.R.: Avoiding Pitfalls When Learning Recursive Theories, in Bajcsy R.(ed.), Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence, Morgan Kaufmann, San Mateo, CA, pp.1050-1057, 1993.

....theory B and a set of observations O, then H 1 [ H 2 need not be consistent with O. This property is the cause of some well known problems when learning multiple predicates or recursive predicates in the explanatory induction setting, cf. De Raedt et al. 1993; Bergadano and Gunetti, 1993; Cameron Jones and Quinlan, 1993] The reason for this is that inconsistencies may arise when H 1 and H 2 can resolve together. Flach s [Flach, 1992] definition of weak induction (from which his later notion of confirmatory induction is derived) is the special case of explanatory induction where only consistency with the ....

R.M. Cameron-Jones and J.R. Quinlan. Avoiding pitfalls when learning recursive theories. In Proceedings of the 13th International Joint Conference on Artificial Intelligence, pages 1050--1055. Morgan Kaufmann, 1993.

No context found.

Cameron-Jones, R.M., and Quinlan, J.R. (1993). Avoiding pitfalls when learning recursive theories (draft). Available by anonymous ftp from

No context found.

R.M. Cameron-Jones and J.R. Quinlan, Avoiding Pitfalls When Learning Recursive Theories, Proceedings Thirteenth International Joint Conference on Artificial Intelligence 1993,1050-1055, Chambery, France.

No context found.

R. M. Cameron-Jones, J.R. Quinlan, 1993. Avoiding Pitfalls When Learning Recursive Theories. Proceedings of the Thirteen International Joint Conference on Artificial Intelligence, Chamb'ery, France, August 28 - September 3, 1993, Vol. 2, pp. 1050-1055.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC